Toward high‐resolution homology modeling of antibody Fv regions and application to antibody–antigen docking

High‐resolution homology models are useful in structure‐based protein engineering applications, especially when a crystallographic structure is unavailable. Here, we report the development and implementation of RosettaAntibody, a protocol for homology modeling of antibody variable regions. The protocol combines comparative modeling of canonical complementarity determining region (CDR) loop conformations and de novo loop modeling of CDR H3 conformation with simultaneous optimization of V_L‐V_H rigid‐body orientation and CDR backbone and side‐chain conformations. The protocol was tested on a benchmark of 54 antibody crystal structures. The median root mean square deviation (rmsd) of the antigen binding pocket comprised of all the CDR residues was 1.5 Å with 80% of the targets having an rmsd lower than 2.0 Å. The median backbone heavy atom global rmsd of the CDR H3 loop prediction was 1.6, 1.9, 2.4, 3.1, and 6.0 Å for very short (4–6 residues), short (7–9), medium (10–11), long (12–14) and very long (17–22) loops, respectively. When the set of ten top‐scoring antibody homology models are used in local ensemble docking to antigen, a moderate‐to‐high accuracy docking prediction was achieved in seven of fifteen targets. This success in computational docking with high‐resolution homology models is encouraging, but challenges still remain in modeling antibody structures for sequences with long H3 loops. This first large‐scale antibody–antigen docking study using homology models reveals the level of “functional accuracy” of these structural models toward protein engineering applications. Proteins 2009; 74:497–514. © 2008 Wiley‐Liss, Inc.     

Remodeling domain interfaces to enhance heterodimer formation.

An anti-p185HER2/anti-CD3 humanized bispecific diabody was previously constructed from two cross-over single-chain Fv in which YH and VL domains of the parent antibodies are present on different polypeptides. Here this diabody is used to evaluate domain interface engineering strategies for enhancing the formation of functional heterodimers over inactive homodimers. A disulfide-stabilized diabody was obtained by introducing two cysteine mutations, VL L46C and VH D101C, at the anti-p185HER2.VL/VH interface. The fraction of recovered diabody that was functional following expression in Escherichia coli was improved for the disulfide-stabilized compared to the parent diabody (> 96% versus 72%), whereas the overall yield was > 60-fold lower. Eleven "knob-into-hole" diabodies were designed by molecular modeling of sterically complementary mutations at the two VL/VH interfaces. Replacements at either interface are sufficient to improve the fraction of functional heterodimer, while maintaining overall recoverable yields and affinity for both antigens close to that of the parent diabody. For example, diabody variant v5 containing the mutations VL Y87A:F98M and VH V37F:L45W at the anti-p185HER2 VL/VH interface was recovered as 92% functional heterodimer while maintaining overall recovered yield within twofold of the parent diabody. The binding affinity of v5 for p185HER2 extracellular domain and T cells is eightfold weaker and twofold stronger than for the parent diabody, respectively. Domain interface remodeling based upon either sterically complementary mutations or interchain disulfide bonds can facilitate the production of a functional diabody heterodimer. This study expands the scope of domain interface engineering by demonstrating the enhanced assembly of proteins interacting via two domain interfaces.     

Improving the productivity of single-chain Fv antibody against c-Met by rearranging the order of its variable domains

Single-chain Fv (scFv) antibody against c-Met is expected to be employed in clinical treatment or imaging of cancer cells owing to the important biological roles of c-Met in the proliferation of malignancies. Here, we show that the productivity of scFv against c-Met in Escherichia coli is significantly influenced by the orientation of its variable domains. We generated anti-c-Met scFv antibodies with two different domain orders (i.e., VL-linker-VH and VHlinker- VL), expressed them in the cytoplasm of E. coli trx/ gor deleted mutant, and compared their specific activities as well as their productivities. Productivity of total and functional anti-c-Met scFv with VH/VL orientation was more than five times higher than that with VL/VH format. Coexpression of DsbC enhanced the yield of soluble amounts of anti-c-Met scFv protein for both constructs. The purified scFv antibodies of the two different formats exhibited almost the same antigen-binding activities. We also compared the productivities and specific activities of anti-c-Met diabodies with VH/VL or VL/VH formats and obtained similar results to the case of scFv antibodies.     

Backbone assignment,secondary structure and Protein A binding of an isolated,human antibody VH domain

Summary Antibody heavy chain variable domains (VH) lacking their light chain domain (VL) partner are prime candidates for the design of minimum-size immunoreagents. To obtain structural information about isolated VH domains, a human VH was labelled with ¹⁵N or doubly labelled with both ¹⁵N and ¹³C and was studied by heteronuclear nuclear magnetic resonance spectroscopy. Most (90%) of the ¹H and ¹⁵N main-chain signals were assigned through two-dimensional TOCSY and NOESY experiments on the unlabelled VH and three-dimensional heteronuclear multiple quantum correlation TOCSY and NOESY experiments on the ¹⁵N-labelled VH. Four short stretches of the polypeptide chain could only be assigned on the basis of three-dimensional HNCA and HN(CO)CA experiments on the ¹³C-/¹⁵N-labelled protein. Long-range interstrand backbone NOEs suggest the presence of two adjacent -sheets formed by altogether nine antiparallel -strands. ³J_H ^NHC coupling constants and the location of slowly exchanging backbone amides support this interpretation. The secondary structure of the isolated VH is identical to that of heavy chain variable domains in intact antibodies, where VH domains are packed against a VL domain. The backbone assignments of the VH made it possible to locate its Protein A binding site. Chemical shift movements after complexing with the IgG binding fragment of Protein A indicate binding through one of the two -sheets of the VH. This -sheet is solvent exposed in intact antibodies. The Protein A binding site obviously differs from that on the Fc portion of immunoglobulins and is unique to members of the human VH_III gene subgroup.Abbreviations CDR complementarity determining region - CHAPS [(cholamidopropyl)-dimethylammonio]-1-propanesulfonate - DQF-COSY double-quantum-filtered correlated spectroscopy - Fab antigen binding antibody fragment - Fc crystallisable antibody fragment - Fv heterodimer of VH and VL - H1 (2, 3) hypervariable loop 1 (2, 3) - IgG immunoglobulin G - NOE nuclear Overhauser effect - NOESY nuclear Overhauser enhancement spectroscopy - HMQC heteronuclear multiple quantum correlation spectroscopy - HSQC heteronuclear single quantum correlation spectroscopy - scFv single chain Fv - TOCSY total correlation spectroscopy - TPPI time-proportional phase incrementation - VH antibody heavy chain variable region - VL antibody light chain variable region. Mutants are denoted by the wild-type amino acid (one-letter code), follwed by the residue number and the new amino acid     

Protein models docking benchmark 2

Structural characterization of protein–protein interactions is essential for our ability to understand life processes. However, only a fraction of known proteins have experimentally determined structures. Such structures provide templates for modeling of a large part of the proteome, where individual proteins can be docked by template‐free or template‐based techniques. Still, the sensitivity of the docking methods to the inherent inaccuracies of protein models, as opposed to the experimentally determined high‐resolution structures, remains largely untested, primarily due to the absence of appropriate benchmark set(s). Structures in such a set should have predefined inaccuracy levels and, at the same time, resemble actual protein models in terms of structural motifs/packing. The set should also be large enough to ensure statistical reliability of the benchmarking results. We present a major update of the previously developed benchmark set of protein models. For each interactor, six models were generated with the model‐to‐native C^α RMSD in the 1 to 6 Å range. The models in the set were generated by a new approach, which corresponds to the actual modeling of new protein structures in the “real case scenario,” as opposed to the previous set, where a significant number of structures were model‐like only. In addition, the larger number of complexes (165 vs. 63 in the previous set) increases the statistical reliability of the benchmarking. We estimated the highest accuracy of the predicted complexes (according to CAPRI criteria), which can be attained using the benchmark structures. The set is available at http://dockground.bioinformatics.ku.edu . Proteins 2015; 83:891–897. © 2015 Wiley Periodicals, Inc.     

Crystal structure to 2.45 A resolution of a monoclonal Fab specific for the Brucella A cell wall polysaccharide antigen.

The atomic structure of an antibody antigen-binding fragment (Fab) at 2.45 A resolution shows that polysaccharide antigen conformation and Fab structure dictated by combinatorial diversity and domain association are responsible for the fine specificity of the Brucella-specific antibody, YsT9.1. It discriminates the Brucella abortus A antigen from the nearly identical Brucella melitensis M antigen by forming a groove-type binding site, lined with tyrosine residues, that accommodates the rodlike A antigen but excludes the kinked structure of the M antigen, as envisioned by a model of the antigen built into the combining site. The variable-heavy (VH) and variable-light (VL) domains are derived from genes closely related to two used in previously solved structures, M603 and R19.9, respectively. These genes combine in YsT9.1 to form an antibody of totally different specificity. Comparison of this X-ray structure with a previously built model of the YsT9.1 combining site based on these homologies highlights the importance of VL:VH association as a determinant of specificity and suggests that small changes at the VL:VH interface, unanticipated in modeling, may cause significant modulation of binding-site properties.     

Recognition of DNA by VH and Fv domains of an IgG anti-poly(dC) antibody with a singly mutated VH domain

Secondary antigen stimulation usually produces IgG antibodies with hypermutated V segments. Studying a strong secondary response to the polynucleotide antigen poly(dC), however, we found a highly selective IgG antibody (mAb dC7) with only one mutation (a conservative Leu to Ileu substitution) throughout the whole VH domain. To investigate the roles of VH and VL domains in selective binding by this mAb, we prepared its VH, VL and single-chain Fv (scFv) fragments. A bacterial expression system produced soluble monomeric V region proteins. CD spectra confirmed that they had the beta-secondary structure expected for Ig domains. Both the scFv and VH fragments bound to single-stranded non-protonated poly(dC) and to ssDNA but not to protonated, more structured poly(dC) or dsDNA. The VL domain alone did not bind to nucleic acids, but VL association modified the VH binding, giving the scFv a 10-fold higher affinity than the VH for poly(dC) and greatly increasing the cytosine-dependent selectivity. Non-ionic interactions were prominent in the Fv reaction with a (dC)( n) sequence. Ionic interactions were revealed in Fv cross-reactions with ssDNA, and were more prominent in binding of either poly(dC) or ssDNA by VH alone, consistent with the lesser base selectivity of the VH. Thus, the Fv and VH alone bind to a single antigen, poly(dC), but mechanistic differences result from additional subsites in the Fv. Generation of a selective IgG with very few CDR mutations in either VH or VL, which was accompanied by IgM antibodies with unmutated V regions, also suggests that nucleic acid binding activity is a property of the B cell repertoire even before immunization.     

Exceptional amyloid beta peptide hydrolyzing activity of nonphysiological immunoglobulin variable domain scaffolds

Nucleophilic sites in the paired variable domains of the light and heavy chains (VL and VH domains) of Ig can catalyze peptide bond hydrolysis. Amyloid beta (Abeta)-binding Igs are under consideration for immunotherapy of Alzheimer disease. We searched for Abeta-hydrolyzing human IgV domains (IgVs) in a library containing a majority of single chain Fv clones mimicking physiological VL-VH-combining sites and minority IgV populations with nonphysiological structures generated by cloning errors. Random screening and covalent selection of phage-displayed IgVs with an electrophilic Abeta analog identified rare IgVs that hydrolyzed Abeta mainly at His14-Gln15. Inhibition of IgV catalysis and irreversible binding by an electrophilic hapten suggested a nucleophilic catalytic mechanism. Structural analysis indicated that the catalytic IgVs are nonphysiological structures, a two domain heterodimeric VL (IgVL2-t) and single domain VL clones with aberrant polypeptide tags (IgVL-t'). The IgVs hydrolyzed Abeta at rates superior to naturally occurring Igs by 3-4 orders of magnitude. Forced pairing of the single domain VL with VH or VL domains resulted in reduced Abeta hydrolysis, suggesting catalysis by the unpaired VL domain.Angstrom level amino acid displacements evident in molecular models of the two domain and unpaired VL domain clones explain alterations of catalytic activity. In view of their superior catalytic activity, the VL domain IgVs may help attain clearance of medically important antigens more efficiently than natural Igs.     

Manual classification strategies in the ECOD database

ECOD (Evolutionary Classification Of protein Domains) is a comprehensive and up‐to‐date protein structure classification database. The majority of new structures released from the PDB (Protein Data Bank) each week already have close homologs in the ECOD hierarchy and thus can be reliably partitioned into domains and classified by software without manual intervention. However, those proteins that lack confidently detectable homologs require careful analysis by experts. Although many bioinformatics resources rely on expert curation to some degree, specific examples of how this curation occurs and in what cases it is necessary are not always described. Here, we illustrate the manual classification strategy in ECOD by example, focusing on two major issues in protein classification: domain partitioning and the relationship between homology and similarity scores. Most examples show recently released and manually classified PDB structures. We discuss multi‐domain proteins, discordance between sequence and structural similarities, difficulties with assessing homology with scores, and integral membrane proteins homologous to soluble proteins. By timely assimilation of newly available structures into its hierarchy, ECOD strives to provide a most accurate and updated view of the protein structure world as a result of combined computational and expert‐driven analysis. Proteins 2015; 83:1238–1251. © 2015 Wiley Periodicals, Inc.     

Examining Variable Domain Orientations in Antigen Receptors Gives Insight into TCR-Like Antibody Design

The variable domains of antibodies and T-Cell receptors (TCRs) share similar structures. Both molecules act as sensors for the immune system but recognise their respective antigens in different ways. Antibodies bind to a diverse set of antigenic shapes whilst TCRs only recognise linear peptides presented by a major histocompatibility complex (MHC). The antigen specificity and affinity of both receptors is determined primarily by the sequence and structure of their complementarity determining regions (CDRs). In antibodies the binding site is also known to be affected by the relative orientation of the variable domains, VH and VL. Here, the corresponding property for TCRs, the Vβ-Vα orientation, is investigated and compared with that of antibodies. We find that TCR and antibody orientations are distinct. General antibody orientations are found to be incompatible with binding to the MHC in a canonical TCR-like mode. Finally, factors that cause the orientation of TCRs and antibodies to be different are investigated. Packing of the long Vα CDR3 in the domain-domain interface is found to be influential. In antibodies, a similar packing affect can be achieved using a bulky residue at IMGT position 50 on the VH domain. Along with IMGT VH 50, other positions are identified that may help to promote a TCR-like orientation in antibodies. These positions should provide useful considerations in the engineering of therapeutic TCR-like antibodies.     

Partial unfolding and refolding for structure refinement: A unified approach of geometric simulations and molecular dynamics

The most successful protein structure prediction methods to date have been template‐based modeling (TBM) or homology modeling, which predicts protein structure based on experimental structures. These high accuracy predictions sometimes retain structural errors due to incorrect templates or a lack of accurate templates in the case of low sequence similarity, making these structures inadequate in drug‐design studies or molecular dynamics simulations. We have developed a new physics based approach to the protein refinement problem by mimicking the mechanism of chaperons that rehabilitate misfolded proteins. The template structure is unfolded by selectively (targeted) pulling on different portions of the protein using the geometric based technique FRODA, and then refolded using hierarchically restrained replica exchange molecular dynamics simulations (hr‐REMD). FRODA unfolding is used to create a diverse set of topologies for surveying near native‐like structures from a template and to provide a set of persistent contacts to be employed during re‐folding. We have tested our approach on 13 previous CASP targets and observed that this method of folding an ensemble of partially unfolded structures, through the hierarchical addition of contact restraints (that is, first local and then nonlocal interactions), leads to a refolding of the structure along with refinement in most cases (12/13). Although this approach yields refined models through advancement in sampling, the task of blind selection of the best refined models still needs to be solved. Overall, the method can be useful for improved sampling for low resolution models where certain of the portions of the structure are incorrectly modeled. Proteins 2015; 83:2279–2292. © 2015 Wiley Periodicals, Inc.     

Prediction of protein domain boundaries from inverse covariances

It has been known even since relatively few structures had been solved that longer protein chains often contain multiple domains, which may fold separately and play the role of reusable functional modules found in many contexts. In many structural biology tasks, in particular structure prediction, it is of great use to be able to identify domains within the structure and analyze these regions separately. However, when using sequence data alone this task has proven exceptionally difficult, with relatively little improvement over the naive method of choosing boundaries based on size distributions of observed domains. The recent significant improvement in contact prediction provides a new source of information for domain prediction. We test several methods for using this information including a kernel smoothing‐based approach and methods based on building alpha‐carbon models and compare performance with a length‐based predictor, a homology search method and four published sequence‐based predictors: DOMCUT, DomPRO, DLP‐SVM, and SCOOBY‐DOmain. We show that the kernel‐smoothing method is significantly better than the other ab initio predictors when both single‐domain and multidomain targets are considered and is not significantly different to the homology‐based method. Considering only multidomain targets the kernel‐smoothing method outperforms all of the published methods except DLP‐SVM. The kernel smoothing method therefore represents a potentially useful improvement to ab initio domain prediction. Proteins 2013. © 2012 Wiley Periodicals, Inc.     

VH-VL orientation prediction for antibody humanization candidate selection: A case study

Antibody humanization describes the procedure of grafting a non-human antibody's complementarity-determining regions, i.e., the variable loop regions that mediate specific interactions with the antigen, onto a β-sheet framework that is representative of the human variable region germline repertoire, thus reducing the number of potentially antigenic epitopes that might trigger an anti-antibody response. The selection criterion for the so-called acceptor frameworks (one for the heavy and one for the light chain variable region) is traditionally based on sequence similarity. Here, we propose a novel approach that selects acceptor frameworks such that the relative orientation of the 2 variable domains in 3D space, and thereby the geometry of the antigen-binding site, is conserved throughout the process of humanization. The methodology relies on a machine learning-based predictor of antibody variable domain orientation that has recently been shown to improve the quality of antibody homology models. Using data from 3 humanization campaigns, we demonstrate that preselecting humanization variants based on the predicted difference in variable domain orientation with regard to the original antibody leads to subsets of variants with a significant improvement in binding affinity.     

A kinetic trap is an intrinsic feature in the folding pathway of single-chain Fv fragments

We have studied the equilibrium unfolding and the kinetics of folding and unfolding of an antibody scFv fragment devoid of cis-prolines. An anti-GCN4 scFv fragment carrying a VL lambda domain, obtained by ribosome display, served as the model system together with an engineered destabilized mutant in VH carrying the R66K exchange. Kinetic and equilibrium unfolding experiments indicate that the VH mutation also affects VL unfolding, possibly by partially destabilizing the interface provided by VH, even though the mutation is distant from the interface. Upon folding of the scFv fragment, a kinetic trap is populated whose escape rate is much faster with the more stable VH domain. The formation of the trap can be avoided if refolding is carried out stepwise, with VH folding first. These results show that antibody scFv fragments do not fold by the much faster independent domain folding, but instead form a kinetically trapped off-pathway intermediate, which slows down folding under native conditions. This intermediate is characterized by premature interaction of the unfolded domains, and particularly involving unfolded VH, independent of proline cis-trans isomerization in VL. This work also implies that VH should be a prime target in engineering well behaving antibody fragments.     

Complete variable region sequences of five homologous high affinity anti-digoxin antibodies

High affinity murine A/J anti-digoxin monoclonal antibodies exhibit diversity in binding specificity for structurally related cardiac glycosides. They utilize several VH and VL genes. Among this diverse set, however, five antibodies share V region amino-terminal sequences that are remarkably homologous. The five antibodies were divided into three subsets based on different fine specificity-binding patterns. Therefore, complete V region sequences were determined by Edman degradation and by nucleotide sequence analysis. The VH region homology among the five antibodies was 84 to 100% and the VL region homology was 89 to 99%. The sequence data are consistent with the use of single (or closely related) VH and VL genes encoding the five antibodies. Four antibodies, derived from the same fusion (40-40, 40-120, 40-140, and 40-160), use identical D, JH2, and JK5 gene segments and identical junctions suggesting that they are clonally related. The fifth antibody (35-20) uses different D and JH1 gene segments but the same JK5 gene segment. All five antibodies share a cross-reactive idiotype. The three antibodies that exhibit the greatest degree of homology (40-40, 40-120, and 40-140) also share indistinguishable antigen-binding patterns as well as private idiotopes not present on the other two antibodies. Antibody 40-160, which has the next most homologous sequence, shares idiotopes with the first set but binds preferentially to different sites on the hapten, whereas antibody 35-20 has the most divergent sequence. In general, the degree of sequence homology among the five antibodies correlates with their hierarchical order based on hapten and idiotype fine specificity patterns.     

The effect of the cosolvent trifluoroethanol on a tryptophan side chain orientation in the hydrophobic core of troponin C

The unique biophysical properties of tryptophan residues have been exploited for decades to monitor protein structure and dynamics using a variety of spectroscopic techniques, such as fluorescence and nuclear magnetic resonance (NMR). We recently designed a tryptophan mutant in the regulatory N‐domain of cardiac troponin C (F77W‐cNTnC) to study the domain orientation of troponin C in muscle fibers using solid‐state NMR. In our previous study, we determined the NMR structure of calcium‐saturated mutant F77W‐V82A‐cNTnC in the presence of 19% 2,2,2‐trifluoroethanol (TFE). TFE is a widely used cosolvent in the biophysical characterization of the solution structures of peptides and proteins. It is generally assumed that the structures are unchanged in the presence of cosolvents at relatively low concentrations, and this has been verified for TFE at the level of the overall secondary and tertiary structure for several calcium regulatory proteins. Here, we present the NMR solution structure of the calcium saturated F77W‐cNTnC in presence of its biological binding partner troponin I peptide (cTnI_144–163) and in the absence of TFE. We have also characterized a panel of six F77W‐cNTnC structures in the presence and absence TFE, cTnI_144–163, and the extra mutation V82A, and used ¹⁹F NMR to characterize the effect of TFE on the F77(5fW) analog. Our results show that although TFE did not perturb the overall protein structure, TFE did induce a change in the orientation of the indole ring of the buried tryptophan side chain from the anticipated position based upon homology with other proteins, highlighting the potential dangers of the use of cosolvents.     

Rearranging the domain order of a diabody-based IgG-like bispecific antibody enhances its antitumor activity and improves its degradation resistance and pharmacokinetics

One approach to creating more beneficial therapeutic antibodies is to develop bispecific antibodies (bsAbs), particularly IgG-like formats with tetravalency, which may provide several advantages such as multivalent binding to each target antigen. Although the effects of configuration and antibody-fragment type on the function of IgG-like bsAbs have been studied, there have been only a few detailed studies of the influence of the variable fragment domain order. Here, we prepared four types of hEx3-scDb-Fc, IgG-like bsAbs, built from a single-chain hEx3-Db (humanized bispecific diabody [bsDb] that targets epidermal growth factor receptor and CD3), to investigate the influence of domain order and fusion manner on the function of a bsDb with an Fc fusion format. Higher cytotoxicities were observed with hEx3-scDb-Fcs with a variable light domain (VL)–variable heavy domain (VH) order (hEx3-scDb-Fc-LHs) compared with a VH–VL order, indicating that differences in the Fc fusion manner do not affect bsDb activity. In addition, flow cytometry suggested that the higher cytotoxicities of hEx3-scDb-Fc-LH may be attributable to structural superiority in cross-linking. Interestingly, enhanced degradation resistance and prolonged in vivo half-life were also observed with hEx3-scDb-Fc-LH. hEx3-scDb-Fc-LH and its IgG2 variant exhibited intense in vivo antitumor effects, suggesting that Fc-mediated effector functions are dispensable for effective anti-tumor activities, which may cause fewer side effects. Our results show that merely rearranging the domain order of IgG-like bsAbs can enhance not only their antitumor activity, but also their degradation resistance and in vivo half-life, and that hEx3-scDb-Fc-LHs are potent candidates for next-generation therapeutic antibodies.     

Structure of the terminal PCP domain of the non‐ribosomal peptide synthetase in teicoplanin biosynthesis

The biosynthesis of the glycopeptide antibiotics, of which teicoplanin and vancomycin are representative members, relies on the combination of non‐ribosomal peptide synthesis and modification of the peptide by cytochrome P450 (Oxy) enzymes while the peptide remains bound to the peptide synthesis machinery. We have structurally characterized the final peptidyl carrier protein domain of the teicoplanin non‐ribosomal peptide synthetase machinery: this domain is believed to mediate the interactions with tailoring Oxy enzymes in addition to its function as a shuttle for intermediates between multiple non‐ribosomal peptide synthetase domains. Using solution state NMR, we have determined structures of this PCP domain in two states, the apo and the post‐translationally modified holo state, both of which conform to a four‐helix bundle assembly. The structures exhibit the same general fold as the majority of known carrier protein structures, in spite of the complex biosynthetic role that PCP domains from the final non‐ribosomal peptide synthetase module must play in glycopeptide antibiotic biosynthesis. These structures thus support the hypothesis that it is subtle rearrangements, rather than dramatic conformational changes, which govern carrier protein interactions and selectivity during non‐ribosomal peptide synthesis. Proteins 2015; 83:711–721. © 2015 Wiley Periodicals, Inc.