Prediction of interactions between cell surface proteins by machine learning

Cells detect changes in their external environments or communicate with each other through proteins on their surfaces. These cell surface proteins form a complicated network of interactions in order to fulfill their functions. The interactions between cell surface proteins are highly dynamic and, thus, challenging to detect using traditional experimental techniques. Here, we tackle this challenge using a computational framework. The primary focus of the framework is to develop new tools to identify interactions between domains in the immunoglobulin (Ig) fold, which is the most abundant domain family in cell surface proteins. These interactions could be formed between ligands and receptors from different cells or between proteins on the same cell surface. In practice, we collected all structural data on Ig domain interactions and transformed them into an interface fragment pair library. A high-dimensional profile can then be constructed from the library for a given pair of query protein sequences. Multiple machine learning models were used to read this profile so that the probability of interaction between the query proteins could be predicted. We tested our models on an experimentally derived dataset that contains 564 cell surface proteins in humans. The cross-validation results show that we can achieve higher than 70% accuracy in identifying the PPIs within this dataset. We then applied this method to a group of 46 cell surface proteins in Caenorhabditis elegans. We screened every possible interaction between these proteins. Many interactions recognized by our machine learning classifiers have been experimentally confirmed in the literature. In conclusion, our computational platform serves as a useful tool to help identify potential new interactions between cell surface proteins in addition to current state-of-the-art experimental techniques. The tool is freely accessible for use by the scientific community. Moreover, the general framework of the machine learning classification can also be extended to study the interactions of proteins in other domain superfamilies.

Molecular Architecture of the Antiophidic Protein DM64 and its Binding Specificity to Myotoxin II From Bothrops asper Venom

DM64 is a toxin-neutralizing serum glycoprotein isolated from Didelphis aurita, an ophiophagous marsupial naturally resistant to snake envenomation. This 64 kDa antitoxin targets myotoxic phospholipases A2, which account for most local tissue damage of viperid snakebites. We investigated the noncovalent complex formed between native DM64 and myotoxin II, a myotoxic phospholipase-like protein from Bothrops asper venom. Analytical ultracentrifugation (AUC) and size exclusion chromatography indicated that DM64 is monomeric in solution and binds equimolar amounts of the toxin. Attempts to crystallize native DM64 for X-ray diffraction were unsuccessful. Obtaining recombinant protein to pursue structural studies was also challenging. Classical molecular modeling techniques were impaired by the lack of templates with more than 25% sequence identity with DM64. An integrative structural biology approach was then applied to generate a three-dimensional model of the inhibitor bound to myotoxin II. I-TASSER individually modeled the five immunoglobulin-like domains of DM64. Distance constraints generated by cross-linking mass spectrometry of the complex guided the docking of DM64 domains to the crystal structure of myotoxin II, using Rosetta. AUC, small-angle X-ray scattering (SAXS), molecular modeling, and molecular dynamics simulations indicated that the DM64-myotoxin II complex is structured, shows flexibility, and has an anisotropic shape. Inter-protein cross-links and limited hydrolysis analyses shed light on the inhibitor's regions involved with toxin interaction, revealing the critical participation of the first, third, and fifth domains of DM64. Our data showed that the fifth domain of DM64 binds to myotoxin II amino-terminal and beta-wing regions. The third domain of the inhibitor acts in a complementary way to the fifth domain. Their binding to these toxin regions presumably precludes dimerization, thus interfering with toxicity, which is related to the quaternary structure of the toxin. The first domain of DM64 interacts with the functional site of the toxin putatively associated with membrane anchorage. We propose that both mechanisms concur to inhibit myotoxin II toxicity by DM64 binding. The present topological characterization of this toxin-antitoxin complex constitutes an essential step toward the rational design of novel peptide-based antivenom therapies targeting snake venom myotoxins.

Structural Insights into N-terminal IgV Domain of BTNL2, a T Cell Inhibitory Molecule, Suggests a Non-canonical Binding Interface for Its Putative Receptors

T cell costimulation is mediated by the interaction of a number of receptors and ligands present on the surface of the T cell and antigen-presenting cell, respectively. Stimulatory or inhibitory signals from these receptor-ligand interactions work in tandem to preserve immune homeostasis. BTNL2 is a type-1 membrane protein that provides inhibitory signal to T cells and plays an important role in several inflammatory and autoimmune diseases. Therefore, manipulation of the molecular interaction of BTNL2 with its putative receptor could provide strategies to restore immune homeostasis in these diseases. Hence, it is imperative to study the structural characteristics of this molecule, which will provide important insights into its function as well. In this study, the membrane-distal ectodomain of murine BTNL2 was expressed in bacteria as inclusion bodies, refolded in vitro and purified for functional and structural characterization. The domain is monomeric in solution as demonstrated by size-exclusion chromatography and analytical ultracentrifugation, and also binds to its putative receptor on naïve B cells and activated T cell subsets. Importantly, for the first time, we report the structure of BTNL2 as determined by solution NMR spectroscopy and also the picosecond-nanosecond timescale backbone dynamics of this domain. The N-terminal ectodomain of BTNL2, which was able to inhibit T cell function as well, exhibits distinctive structural features. The N-terminal ectodomain of BTNL2 has a significantly reduced surface area in the front sheet due to the non-canonical conformation of the CC' loop, which provides important insights into the recognition of its presently unknown binding partner.

Structural basis for the recognition of nectin-like protein-5 by the human-activating immune receptor, DNAM-1

Nectin and nectin-like (Necl) adhesion molecules are broadly overexpressed in a wide range of cancers. By binding to these adhesion molecules, the immunoreceptors DNAX accessory molecule-1 (DNAM-1), CD96 molecule (CD96), and T-cell immunoreceptor with Ig and ITIM domains (TIGIT) play a crucial role in regulating the anticancer activities of immune effector cells. However, within this axis, it remains unclear how DNAM-1 recognizes its cognate ligands. Here, we determined the structure of human DNAM-1 in complex with nectin-like protein-5 (Necl-5) at 2.8 Å resolution. Unexpectedly, we found that the two extracellular domains (D1-D2) of DNAM-1 adopt an unconventional "collapsed" arrangement that is markedly distinct from those in other immunoglobulin-based immunoreceptors. The DNAM-1/Necl-5 interaction was underpinned by conserved lock-and-key motifs located within their respective D1 domains, but also included a distinct interface derived from DNAM-1 D2. Mutation of the signature DNAM-1 "key" motif within the D1 domain attenuated Necl-5 binding and natural killer cell-mediated cytotoxicity. Altogether, our results have implications for understanding the binding mode of an immune receptor family that is emerging as a viable candidate for cancer immunotherapy.

Human cytomegalovirus-specific T-cell receptor engineered for high affinity and soluble expression using mammalian cell display

T-cell receptors (TCR) have considerable potential as therapeutics and antibody-like reagents to monitor disease progression and vaccine efficacy. Whereas antibodies recognize only secreted and surface-bound proteins, TCRs recognize otherwise inaccessible disease-associated intracellular proteins when they are presented as processed peptides bound to major histocompatibility complexes (pMHC). TCRs have been primarily explored for cancer therapy applications but could also target infectious diseases such as cytomegalovirus (CMV). However, TCRs are more difficult to express and engineer than antibodies, and advanced methods are needed to enable their widespread use. Here, we engineered the human CMV-specific TCR RA14 for high-affinity and robust soluble expression. To achieve this, we adapted our previously reported mammalian display system to present TCR extracellular domains and used this to screen CDR3 libraries for clones with increased pMHC affinity. After three rounds of selection, characterized clones retained peptide specificity and activation when expressed on the surface of human Jurkat T cells. We obtained high yields of soluble, monomeric protein by fusing the TCR extracellular domains to antibody hinge and Fc constant regions, adding a stabilizing disulfide bond between the constant domains and disrupting predicted glycosylation sites. One variant exhibited 50 nm affinity for its cognate pMHC, as measured by surface plasmon resonance, and specifically stained cells presenting this pMHC. Our work has identified a human TCR with high affinity for the immunodominant CMV peptide and offers a new strategy to rapidly engineer soluble TCRs for biomedical applications.

A single residue switch reveals principles of antibody domain integrity

Despite their importance for antibody architecture and design, the principles governing antibody domain stability are still not understood in sufficient detail. Here, to address this question, we chose a domain from the invariant part of IgG, the C_H2 domain. We found that compared with other Ig domains, the isolated C_H2 domain is a surprisingly unstable monomer, exhibiting a melting temperature of ∼44 °C. We further show that the presence of an additional C-terminal lysine in a C_H2 variant substantially increases the melting temperature by ∼14 °C relative to C_H2 WT. To explore the molecular mechanism of this effect, we employed biophysical approaches to probe structural features of C_H2. The results revealed that Lys¹⁰¹ is key for the formation of three secondary structure elements: the very C-terminal β-strand and two adjacent α-helices. We also noted that a dipole interaction between Lys¹⁰¹ and the nearby α-helix, is important for stabilizing the C_H2 architecture by protecting the hydrophobic core. Interestingly, this interaction between the α-helix and C-terminal charged residues is highly conserved in antibody domains, suggesting that it represents a general mechanism for maintaining their integrity. We conclude that the observed interactions involving terminal residues have practical applications for defining domain boundaries in the development of antibody therapeutics and diagnostics.

Crystal structure of B-cell co-receptor CD19 in complex with antibody B43 reveals an unexpected fold

CD19 is a transmembrane protein expressed on malignant B cells, but not in other lineages or other tissues, which makes it an attractive target for monoclonal antibody-mediated immunotherapy. Anti-CD19 antibody B43 was utilized in a bispecific T-cell engager (BiTE) blinatumomab that demonstrated potency for the treatment of relapsed acute lymphoblastic leukemia. To gain insight into the mechanism of action of the antibody, the crystal structure of B43 Fab was determined in complex with CD19 and in the unbound form. The structure revealed the binding epitope, explained the lack of cross-reactivity toward non-human species, and suggested the key-and-lock mechanism of antigen recognition. Most unexpectedly, the structure revealed a unique molecular topology of CD19. Rather than a tandem of c-type immunoglobulin folds predicted from the amino acid sequence, the extracellular domain of CD19 exhibits an elongated β-sandwich formed by two immunoglobulin folds by swapping their C-terminal halves. This is the first structure of CD19, which has no sequence homologs.

A peptide extension dictates IgM assembly

Professional secretory cells can produce large amounts of high-quality complex molecules, including IgM antibodies. Owing to their multivalency, polymeric IgM antibodies provide an efficient first-line of defense against pathogens. To decipher the mechanisms of IgM assembly, we investigated its biosynthesis in living cells and faithfully reconstituted the underlying processes in vitro. We find that a conserved peptide extension at the C-terminal end of the IgM heavy (Ig-μ) chains, termed the tailpiece, is necessary and sufficient to establish the correct geometry. Alanine scanning revealed that hydrophobic amino acids in the first half of the tailpiece contain essential information for generating the correct topology. Assembly is triggered by the formation of a disulfide bond linking two tailpieces. This induces conformational changes in the tailpiece and the adjacent domain, which drive further polymerization. Thus, the biogenesis of large and topologically challenging IgM complexes is dictated by a local conformational switch in a peptide extension.

Role of the disulfide bond in stabilizing and folding of the fimbrial protein DraE from uropathogenic Escherichia coli

Dr fimbriae are homopolymeric adhesive organelles of uropathogenic Escherichia coli composed of DraE subunits, responsible for the attachment to host cells. These structures are characterized by enormously high stability resulting from the structural properties of an Ig-like fold of DraE. One feature of DraE and other fimbrial subunits that makes them peculiar among Ig-like domain-containing proteins is a conserved disulfide bond that joins their A and B strands. Here, we investigated how this disulfide bond affects the stability and folding/unfolding pathway of DraE. We found that the disulfide bond stabilizes self-complemented DraE (DraE-sc) by ∼50 kJ mol⁻¹ in an exclusively thermodynamic manner, i.e. by lowering the free energy of the native state and with almost no effect on the free energy of the transition state. This finding was confirmed by experimentally determined folding and unfolding rate constants of DraE-sc and a disulfide bond-lacking DraE-sc variant. Although the folding of both proteins exhibited similar kinetics, the unfolding rate constant changed upon deletion of the disulfide bond by 10 orders of magnitude, from ∼10⁻¹⁷ s⁻¹ to 10⁻⁷ s⁻¹. Molecular simulations revealed that unfolding of the disulfide bond-lacking variant is initiated by strands A or G and that disulfide bond-mediated joining of strand A to the core strand B cooperatively stabilizes the whole protein. We also show that the disulfide bond in DraE is recognized by the DraB chaperone, indicating a mechanism that precludes the incorporation of less stable, non-oxidized DraE forms into the fimbriae.

Recognition of nectin-2 by the natural killer cell receptor T cell immunoglobulin and ITIM domain (TIGIT)

T cell immunoglobulin and ITIM domain (TIGIT) is an inhibitory receptor expressed on the surface of natural killer (NK) cells. TIGIT recognizes nectin and nectin-like adhesion molecules and thus plays a critical role in the innate immune response to malignant transformation. Although the TIGIT nectin-like protein-5 (necl-5) interaction is well understood, how TIGIT engages nectin-2, a receptor that is broadly over-expressed in breast and ovarian cancer, remains unknown. Here, we show that TIGIT bound to the immunoglobulin domain of nectin-2 that is most distal from the membrane with an affinity of 6 μm, which was moderately lower than the affinity observed for the TIGIT/necl-5 interaction (3.2 μm). The TIGIT/nectin-2 binding disrupted pre-assembled nectin-2 oligomers, suggesting that receptor-ligand and ligand-ligand associations are mutually exclusive events. Indeed, the crystal structure of TIGIT bound to the first immunoglobulin domain of nectin-2 indicated that the receptor and ligand dock using the same molecular surface and a conserved "lock and key" binding motifs previously observed to mediate nectin/nectin homotypic interactions as well as TIGIT/necl-5 recognition. Using a mutagenesis approach, we dissected the energetic basis for the TIGIT/nectin-2 interaction and revealed that an "aromatic key" of nectin-2 is critical for this interaction, whereas variations in the lock were tolerated. Moreover, we found that the C-C' loop of the ligand dictates the TIGIT binding hierarchy. Altogether, these findings broaden our understanding of nectin/nectin receptor interactions and have implications for better understanding the molecular basis for autoimmune disease and cancer.

Aggregation of Full-length Immunoglobulin Light Chains from Systemic Light Chain Amyloidosis (AL) Patients Is Remodeled by Epigallocatechin-3-gallate

Intervention into amyloid deposition with anti-amyloid agents like the polyphenol epigallocatechin-3-gallate (EGCG) is emerging as an experimental secondary treatment strategy in systemic light chain amyloidosis (AL). In both AL and multiple myeloma (MM), soluble immunoglobulin light chains (LC) are produced by clonal plasma cells, but only in AL do they form amyloid deposits in vivo We investigated the amyloid formation of patient-derived LC and their susceptibility to EGCG in vitro to probe commonalities and systematic differences in their assembly mechanisms. We isolated nine LC from the urine of AL and MM patients. We quantified their thermodynamic stabilities and monitored their aggregation under physiological conditions by thioflavin T fluorescence, light scattering, SDS stability, and atomic force microscopy. LC from all patients formed amyloid-like aggregates, albeit with individually different kinetics. LC existed as dimers, ∼50% of which were linked by disulfide bridges. Our results suggest that cleavage into LC monomers is required for efficient amyloid formation. The kinetics of AL LC displayed a transition point in concentration dependence, which MM LC lacked. The lack of concentration dependence of MM LC aggregation kinetics suggests that conformational change of the light chain is rate-limiting for these proteins. Aggregation kinetics displayed two distinct phases, which corresponded to the formation of oligomers and amyloid fibrils, respectively. EGCG specifically inhibited the second aggregation phase and induced the formation of SDS-stable, non-amyloid LC aggregates. Our data suggest that EGCG intervention does not depend on the individual LC sequence and is similar to the mechanism observed for amyloid-β and α-synuclein.

The Antibody Light-Chain Linker Is Important for Domain Stability and Amyloid Formation

The association of light chains (LCs) and heavy chains is the basis for functional antibodies that are essential for adaptive immune responses. However, in some cases, LCs and especially fragments consisting of the LC variable (VL) domain are pathologically deposited in fatal aggregation diseases. The two domains of the LC are connected by a highly conserved linker. We show here that, unexpectedly, the linker residue Arg108 affects the conformational stability and folding of both VLκ and LC constant (CLκ) domains. Interestingly, the extension of VL by Arg108 results in its resistance to amyloid formation, which suggests that the nature of the truncation of the LC plays a crucial role in disease progression. Increased solvation due to the exposed charged C-terminal Arg108 residue explains its stabilizing effects on the VL domain. For the CL domain, the interaction of N-terminal loop residues with Arg108 is important for the integrity of the domain, as the disruption of this interaction results in fluctuation, partial opening of the protein's interior and the exposure of hydrophobic residues that destabilize the domain. This establishes new principles for antibody domain architecture and amyloidogenicity.

Mutations can cause light chains to be too stable or too unstable to form amyloid fibrils

Light chain (AL) amyloidosis is an incurable human disease, where the amyloid precursor is a misfolding-prone immunoglobulin light-chain. Here, we identify the role of somatic mutations in the structure, stability and in vitro fibril formation for an amyloidogenic AL-12 protein by restoring four nonconservative mutations to their germline (wild-type) sequence. The single restorative mutations do not affect significantly the native structure, the unfolding pathway, and the reversibility of the protein. However, certain mutations either decrease (H32Y and H70D) or increase (R65S and Q96Y) the protein thermal stability. Interestingly, the most and the least stable mutants, Q96Y and H32Y, do not form amyloid fibrils under physiological conditions. Thus, Q96 and H32 are key residues for AL-12 stability and fibril formation and restoring them to the wild-type residues preclude amyloid formation. The mutants whose equilibrium is shifted to either the native or unfolded states barely sample transient partially folded states, and therefore do not form fibrils. These results agree with previous observations by our laboratory and others that amyloid formation occurs because of the sampling of partially folded states found within the unfolding transition (Blancas-Mejia and Ramirez-Alvarado, Ann Rev Biochem 2013;82:745-774). Here we provide a new insight on the AL amyloidosis mechanism by demonstrating that AL-12 does not follow the established thermodynamic hypothesis of amyloid formation. In this hypothesis, thermodynamically unstable proteins are more prone to amyloid formation. Here we show that within a thermal stability range, the most stable protein in this study is the most amyloidogenic protein.

The response of Greek key proteins to changes in connectivity depends on the nature of their secondary structure

What governs the balance between connectivity and topology in regulating the mechanism of protein folding? We use circular permutation to vary the order of the helices in the all-α Greek key protein FADD (Fas-associated death domain) to investigate this question. Unlike all-β Greek key proteins, where changes in the order of secondary structure cause a shift in the folding nucleus, the position of the nucleus in FADD is unchanged, even when permutation reduces the complexity significantly. We suggest that this is because local helical contacts are so dominant that permutation has little effect on the entropic cost of forming the folding nucleus whereas, in all-β Greek key proteins, all interactions in the nucleus are long range. Thus, the type of secondary structure modulates the sensitivity of proteins to changes in connectivity.

Nectin family of cell-adhesion molecules: structural and molecular aspects of function and specificity

Cell-cell adhesive processes are central to the physiology of multicellular organisms. A number of cell surface molecules contribute to cell-cell adhesion, and the dysfunction of adhesive processes underlies numerous developmental defects and inherited diseases. The nectins, a family of four immunoglobulin superfamily members (nectin-1 to -4), interact through their extracellular domains to support cell-cell adhesion. While both homophilic and heterophilic interactions among the nectins are implicated in cell-cell adhesion, cell-based and biochemical studies suggest heterophilic interactions are stronger than homophilic interactions and control a range of physiological processes. In addition to interactions within the nectin family, heterophilic associations with nectin-like molecules, immune receptors, and viral glycoproteins support a wide range of biological functions, including immune modulation, cancer progression, host-pathogen interactions and immune evasion. We review current structural and molecular knowledge of nectin recognition processes, with a focus on the biochemical and biophysical determinants of affinity and selectivity that drive distinct nectin associations. These proteins and interactions are discussed as potential targets for immunotherapy.

Teaching the structure of immunoglobulins by molecular visualization and SDS-PAGE analysis

This laboratory class combines molecular visualization and laboratory experimentation to teach the structure of the immunoglobulins (Ig). In the first part of the class, the three-dimensional structures of the human IgG and IgM molecules available through the RCSB PDB database are visualized using freely available software. In the second part, IgG and IgM are studied using electrophoretic methods. Through SDS-PAGE analysis under reducing conditions, the students determine the number and molecular masses of the polypeptide chains, while through SDS-PAGE under nonreducing conditions, the students assess the oligomerization of these Ig molecules. The aims of this class are to expand upon the knowledge and understanding of the Ig structure that the students have gained from classroom lectures. The combination of this molecular visualization of the Ig molecules and the SDS-PAGE experimentation ensures variety in the teaching techniques, while the implication of the Ig molecules in human disease promotes interest for biomedical students.

Hydrophobic core in domains of immunoglobulin-like fold

This work analyzes proteins which contain an immunoglobulin fold, focusing on their hydrophobic core structure. The "fuzzy oil drop" model was used to measure the regularity of hydrophobicity distribution in globular domains belonging to proteins which exhibit the above-mentioned fold. Light-chain IgG domains are found to frequently contain regular hydrophobic cores, unlike the corresponding heavy-chain domains. Enzymes and DNA binding proteins present in the data-set are found to exhibit poor accordance with the hydrophobic core model.