Ontogeny of Recognition Specificity and Functionality for the Broadly Neutralizing Anti-HIV Antibody 4E10

The process of antibody ontogeny typically improves affinity, on-rate, and thermostability, narrows polyspecificity, and rigidifies the combining site to the conformer optimal for binding from the broader ensemble accessible to the precursor. However, many broadly-neutralizing anti-HIV antibodies incorporate unusual structural elements and recognition specificities or properties that often lead to autoreactivity. The ontogeny of 4E10, an autoreactive antibody with unexpected combining site flexibility, was delineated through structural and biophysical comparisons of the mature antibody with multiple potential precursors. 4E10 gained affinity primarily by off-rate enhancement through a small number of mutations to a highly conserved recognition surface. Controverting the conventional paradigm, the combining site gained flexibility and autoreactivity during ontogeny, while losing thermostability, though polyspecificity was unaffected. Details of the recognition mechanism, including inferred global effects due to 4E10 binding, suggest that neutralization by 4E10 may involve mechanisms beyond simply binding, also requiring the ability of the antibody to induce conformational changes distant from its binding site. 4E10 is, therefore, unlikely to be re-elicited by conventional vaccination strategies.


Introduction
An effective HIV vaccine will likely need to elicit broadlyneutralizing antibodies (bnAbs) that target the viral envelope protein (Env) as part of a protective immune response [1][2][3][4][5][6]. Envderived and reverse-engineered immunogen-based vaccines, however, have consistently failed to elicit bnAbs. Possible explanations include that: (1) immunogens may be unable to bind germlineencoded precursors (GEPs) of bnAbs with sufficient affinity to initiate B cell activation and affinity maturation, which has a ,micromolar threshold [7][8][9]; (2) rearranged V H and V L genes compatible with the development of bnAbs may not be common in human or animal model vaccinee GEP repertoires; (3) some bnAbs are autoreactive, which hinders their elicitation through self-tolerance mechanisms; (4) the unusual characteristics inherent to bnAbs, such as long complementarity determining regions (CDRs), functionally-required polyspecificity, and a high degree of somatic mutation (typically observed in Abs elicited in response to chronic infections, including bnAbs), may not be easily achieved through conventional vaccination strategies; (5) imperfect immunogens may elicit off-target (non-neutralizing or non-Env) or humoral responses with limited breadth; and, finally, (6) neutralization mechanisms may involve complexities beyond simply binding a particular epitope on Env (e.g., inducing specific conformational changes), which may be difficult to recapitulate, since selective expansion of particular B cell clones is based solely on BCR binding properties, not higher-order functionalities [10,11].
The bnAb 4E10, the focus of our studies, has a conserved, linear epitope (core epitope: 671 NWF D / N IT 676 ) immediately adjacent to the viral membrane in the Env gp41 subunit membrane proximal external region (MPER) [12,13]. While 4E10 displays admirable breadth [14], has been the target of a successful design effort to reverse-engineer tight-binding immunogens [15], has recognizable GEPs present at finite frequencies in human naïve repertoires [16], and arguably does not require a high degree of polyspecificity to neutralize HIV [17,18], its viability as a vaccine target is hampered by limited potency, demonstrated autoreactivity and exceptional combining site flexibility [17,19,20]. The neutralization mechanism of 4E10 also has not been clearly defined and may involve higher order effects [17,21,22]. The ontogeny of 4E10, therefore, must be elucidated in order to understand how these properties were acquired and to what degree they impose constraints that might hinder re-elicitation by vaccination.
Mutations acquired during Ab maturation occur preferentially in the CDRs, which make up the six loops (CDRs 1, 2, and 3 on the heavy (HCDRs) and light (LCDRs) chains) comprising the combining site [23,24]. CDRs are responsible for the majority of direct contacts to an antigen, as opposed to the intervening framework regions (FWRs), which form the immunoglobulin bsheet structure stabilizing the combining site, helping define CDR loop conformations. While CDR mutations are typically thought to more directly affect antigen binding and neutralization, bnAbs consistently depend on FWR substitutions to a surprising degree, though 4E10 is an exception to this exception [4]. bnAbs are notorious for their high degree of somatic hypermutation, the product of a long process of affinity maturation against a rapidly mutating virus during a persistent, chronic infection [25]. While typical affinity-matured Abs have acquired 15 to 20 V H mutations, bnAbs accumulate up to 100 V H mutations [4]. These mutations are crucial because reversion to germline sequences drastically reduces epitope affinity and neutralization potency and breadth. In many cases, bnAb GEPs are unable to bind Env, though the actual eliciting isolate may not be known [26][27][28][29][30][31]. In addition, bnAbs can contain extraordinarily long HCDR3s, up to 33 residues long versus an average of 13 for non-HIV bnAbs [32,33]. Using phylogenetically-inferred GEP sequences [16,34] (Fig. 1), 4E10 has acquired between 33 and 35 mutations during maturation, 20 to 22 in V H , depending on gene segment selection, and 13 in V L , and has an HCDR3 22 residues long, values at the less exceptional end of the bnAb spectrum and not unheard of for conventional Abs.
Affinity-matured Abs display univalent equilibrium binding constants (K D ) for their cognate antigens, typically ranging from 10 26 to 10 210 M, that are orders-of-magnitude stronger than their GEPs [35]. Multiple approaches, including computational analyses and biophysical comparisons of affinity-matured Abs and their associated GEPs, have generated a consensus model for the molecular mechanism of affinity maturation [36][37][38][39][40][41][42][43][44][45][46][47], perhaps better understood as binding optimization, that traces its roots back to Pauling [48]. In the consensus model, the antigen specificities of the naïve, germline-encoded repertoire, while diverse and extensive, are further extended by encoding a high degree of polyspecificity into GEPs. This is accomplished partly through increased combining site plasticity in GEPs, more formally stated as the ability of GEP CDRs to dynamically sample a broader ensemble of structural conformers. In response to immunogen stimulation, Ab binding properties are iteratively optimized through cycles of somatic hypermutation and selection, resulting in improved binding affinities, kinetics and thermodynamics. While mutations have been observed to improve or add direct contacts to antigen, typically improving enthalpies of interaction and off-rates (k d ), a majority of measurably favorable mutations do not directly contact antigen. These mutations indirectly optimize binding by: (1) increasing shape complementarity Author Summary 4E10 is an antibody that neutralizes a broad variety of HIV strains. However, 4E10 is uncommon in infected patients and has not been successfully elicited by any vaccine approach attempted. Hurdles to re-eliciting 4E10 include the accumulation of many mutations during development, demonstrated reactivity against host proteins and significant structural flexibility. Lacking a confirmed sequence for precursors of 4E10, we studied the recognition and biophysical properties of an ensemble of eight of the likeliest candidates. Surprisingly, 4E10 gained host reactivity and structural flexibility, but lost stability during development when compared to candidate precursors. However, recognition of HIV was remarkably conserved, despite a considerable improvement in binding. Since these results run counter to those expected from conventional vaccination protocols, 4E10 is unlikely to serve as the basis of a useful HIV vaccine. domains are shown, with CDRs indicated by a blue overscore. Predicted somatic mutations are colored red in the 4E10 sequence, and the corresponding unmutated GEP residues are shown below in grey (unchanged positions are not shown for clarity). All GEP V L domains are comprised of the IGKV3-20*01/IGKJ1*01 gene segment combination. Each GEP V H domain comprises the IGHV1-69*06 V gene segment plus one of six D gene segments (listed to the left of the corresponding GEP in the blue boxed field), and either the IGHJ4*02 (resulting sequence differences shown in purple and bolded) or IGHJ1*01 (resulting sequence difference shown in blue and bolded) gene segment, yielding an ensemble of 12 GEPs in toto. GEP shorthand numbering is shown in grey (GEPs 1 to 6) and blue (GEPs 7 to 12) beside the corresponding D plus J gene segment combination sequence differences from 4E10, occurring in HCDR3. Inset (lower right): HCDR3 sequences from candidate 4E10 GEPs, determined through deep sequencing of naïve B cell germline IgH genes from four uninfected individuals [16], show the degree of variability seen in potential 4E10 precursors present in naïve repertoires. Each germline rearrangement uses the IGHV1-69 and IGJH1 or IGJH4 gene segments. Amino acids in red designate sequence differences between GEP and 4E10. The number of nucleotide changes needed to achieve these somatic mutations is 16  between paratopes and epitopes through more global effects on structure; (2) increasing antibody stability, typically measured as solution thermostabilities (T m ), thus compensating for deleterious effects of other mutations that improve affinity; and (3) structurally rigidifying the combining site conformer optimal for binding antigen from the accessible ensemble. Rigidifying the combining site can affect measured interaction parameters in different ways depending on the mechanism of binding. The two mechanistic extremes are known as ''conformer-selection'' and ''instructive-encounter'', or ''induced-fit'', binding. In conformerselection mode, binding does not occur until the compatible conformer is adopted. Rigidification of the binding site through mutation then typically improves entropies of interaction and onrates (k a ). In instructive-encounter mode, initial binding occurs to sub-optimal conformers which affects the rate of interchange with the optimal conformer. Rigidification of the binding site through mutation then typically improves affinity through changes distributed over k d and k a . However, the consensus of studies of binding proteins and enzymes suggests that conformer selection is the preeminent recognition mechanism [49].
Surprisingly, comparisons of the bound and unbound structures of 4E10 revealed that this affinity-matured bnAb incorporates considerable HCDR conformational flexibility [17], in excess of what has been observed in most other antibodies, mature or GEP, suggesting that the ontogeny of 4E10 may be an exception to the consensus model and may pose unique challenges as a vaccine target. In order to fully understand the ontogeny of this unique bnAb and consequences for vaccine development, we characterized the unbound and complex crystal structures, and the functional and binding properties, of 4E10 and an ensemble of its most likely GEPs. GEPs showed detectable, but extremely weak, binding to soluble Env gp140s and extremely limited neutralization potency, though some reverse engineered epitope-scaffolds (ESs) showed robust GEP affinities, well above the B cell activation threshold. 4E10 and GEP paratopes displayed a remarkable degree of structural conservation in the antigen-bound state, with little improvement in overall shape complementarity. Multi-log improvements in affinity for ESs were the result of improved off-rates or combined improvements in onand off-rates, with a small number of enhanced contacts to antigen observed in the crystal structures. However, minimal mutations of GEP sequences to include these enhanced direct contacts only marginally increases affinity. FWR mutations had little discernable effect on global or local structure. Controverting the consensus model of ontogeny, 4E10 thermostability was appreciably worse than its GEPs; while 4E10 and GEPs displayed similarly constrained V H /V L interdomain movements upon binding, 4E10 maturation involved negligible combining site rigidification, with both 4E10 and GEP HCDRs sampling extensive conformer ensembles. The narrowing of polyspecificity assumed to concur with maturation was not observed with 4E10, as both 4E10 and its GEPs showed similar patterns of limited polyspecificity to a phage-displayed human peptidome (Phage Immunoprecipitation Sequencing; PhIP-Seq) [50]. While 4E10 is demonstrably autoreactive, GEPs exhibited a distinct profile of autoantigen recognition by PhIP-Seq. When combined, these results inform efforts to re-elicit 4E10 by vaccination and its mechanism of neutralization.

Prediction of an ensemble of likeliest GEP sequences
Lacking access to the original donor, identification of a single GEP sequence with high confidence for many bnAbs, including 4E10, is complicated by extensive editing and TdT N-nucleotide insertion during rearrangement [26], leading to our decision to study an ensemble of the 12 likeliest candidates (Fig. 1). Due to the prediction that the sequence differences introduced by alternate heavy chain J gene segments may not affect any discernable GEP property, the initial ensemble was limited to eight GEPs (IGHJ4*02 paired with all six D segments plus IGHJ1*01 paired with IGHD1-1*01 and IGHD6-19*01), with the intention of generating additional GEPs if the IGHJ4*02/ IGHJ1*01 substitution exhibited any differences in structure or binding properties on the IGHD1-1*01 or IGHD6-19*01 backgrounds. These eight GEPs also recapitulated some of the variability seen in potential 4E10 GEPs identified by deep sequencing of uninfected individuals ( Fig. 1) [16], providing an additional justification for studying the ensemble. However, GEPs in our ensemble differed from one another by no more than four mutations, though the mutations were quite nonconservative.

GEP protein production and validation
GEPs were engineered as cleavable, single-chain, variable domain cassette (Fv; V H +V L ) constructs to ease expression, analysis, and crystallization and to prevent monobody-diabody interchange, following protocols developed for 4E10 [18]. The prior study confirmed that these Fv constructs retained the structural and binding properties of Fab fragments of intact 4E10. All eight GEP Fvs expressed at high levels as bacterial inclusion bodies and, in all but one case (GEP 5), were refolded in vitro with what in our experience were exceptionally high efficiencies of 20 to 40%. GEP 5 was not included in subsequent experiments because its extremely poor in vitro refolding efficiency suggested that this was not a viable pairing. GEP constructs were stable and monodisperse in solution, running exclusively as monomers by size exclusion chromatography (SEC). Reduced/non-reduced PAGE analysis of GEPs confirmed purity and proper disulfide bond formation. GEP T m values ( Fig. 2A), determined by circular dichroism (CD) spectroscopy as previously described [18], narrowly ranged from 64.2uC to 67.0uC, showing that GEP Fvs were well folded, but had even higher T m values than 4E10 Fv (52.8uC).

GEP neutralization potencies were dramatically reduced
The neutralization potency of GEPs was tested against clade A (Q461.d1, Q461.e2 [51]) and B (SF162 [52]) HIV-1 isolates using standard TZM-bl assays (Fig. 2B) [53]. Overall, GEP Fv potencies were markedly reduced in comparison with 4E10 Fv. GEPs displayed only very weak and likely insignificant neutralization potencies, though with a trend of greater effect against the clade A isolates, particularly the neutralization-sensitive strain Q461.d1, and with GEP 1 and GEP 7 showing marginally better potencies across tested isolates.  Table 1 GEPs bound Env proteins and engineered antigens, but with reduced affinities In order to characterize the change in binding properties during 4E10 ontogeny, the binding of 4E10 and GEP Fvs to three (SF162, SF162 K160N and Q461.e2) soluble Env gp140 trimers (gp140 3 ) [54] and four engineered 4E10 ES immunogens (T72, T93, T117 and T344) [15,55] was evaluated by surface plasmon resonance (SPR) interaction analysis (Figs. 2C, 2D, Figs. 3-7, Table 1). Isolates and ESs were selected to span a range of binding properties, where previous studies had shown 4E10 bound a free peptide spanning its linear epitope with a K D of 12 nM, SF162 gp140 3 with a K D of 98 nM, and ESs with K D s of either #10 picomolar (T117) or ,100 picomolar (T72, T93 and T344) [15,18,55]. All seven GEPs showed unquantifiably weak, but detectable binding to chip-coupled clade A (Q461.e2) and clade B (SF162, SF162 K160N ) gp140 3 and T72, T93, and T344 in qualitative SPR analyses, with K D s all estimated to be well above the ,micromolar B cell activation threshold. Quantitative SPR analyses of GEPs binding to T117 showed K D s ranging from the low nanomolar to low micromolar range. 4E10 interactions with ESs ranged from 100-to 10,000-fold stronger than GEPs, which was qualitatively consistent with the observed difference in 4E10 versus GEP interactions with gp140 3 . Since the GEP/gp140 3 interactions were too weak to quantitate, peptide binding studies were not performed on the expectation that they would also be too weak to measure accurately. Five GEPs (1, 2, 3, 7 and 8) bind T117 with nearly identical behavior, including GEP 1 and GEP 7, which differ only by alternate J segment utilization, showing that the two incorporated mutations did not affect binding, so no further IGHJ1*01 GEPs were produced for study. GEP 4 and GEP 6, which both incorporate differences from G96H in 4E10 (A or V), showed approximately 10-fold (GEP 4) or 100-fold (GEP 6) reductions in affinity relative to the cluster of other GEPs. Kinetically, the five clustering GEPs (1, 2, 3, 7 and 8) showed affinity reductions for T117 relative to 4E10 overwhelmingly through faster off-rates (k d s). GEPs 4 and 6, in addition to comparable increases in k d , also showed progressive reductions in on-rates (k a s).

4E10-and GEP-ES complex structures show binding site conservation
In order to shed light on potential structural differences accounting for reduced GEP binding affinities, crystal structures of GEP 1, 2 and 7 in complex with T117 were determined at resolution values of 2.9 Å , 1.8 Å , and 3.1 Å respectively, rebuilt and refined with good statistics ( Table 2), and compared to two reference 4E10/antigen complex structures: 4E10 bound to an epitope peptide (2FX7.pdb [12]) or a related ES, T88 (3LH2.pdb [15]). Superpositions showed that almost all direct contacts to the core epitope (NWFDIT) and epitope conformation are conserved between 4E10 and GEP complexes to a remarkable degree (   Table 1. doi:10.1371/journal.ppat.1004403.g004 mutations affect sequence positions making direct contacts to the core epitope: Y/K32L, S/Q93L, T/S31H, I/V51H, F/L54H and T/I56H. However, residues at two of these positions (93L, 51H) contacted the epitope solely through main-chain interactions, which were not affected by the mutated side-chains. Replacements at three other positions (31H, 54H, 56H) nearly perfectly recapitulated contacts, also conserving the interface. The 4E10 and GEP paratope/epitope interface is largely hydrophobic, thus binding is mainly mediated by Van der Waals contacts and desolvation entropy, which is conserved through equivalent positioning of nonpolar groups. This was reflected in the close concordance between the surface areas (SA) buried in the complexes by core epitope and the corresponding shape complementarity (Sc) [56] between Ab and core epitope. Only two somatic mutations were predicted to contribute to a stronger binding interaction. The Y/ K32L mutation replaces a hydrogen bond involving the tyrosine hydroxyl with a salt bridge. The P/L95H mutation, involving noncontacting residues, restructured LCDR3 to reposition the sidechain of the conserved serine at 94L from a non-contacting position in GEPs to one contributing a hydrogen bond in 4E10 complex structures. However, the Y/K32L-P/L95H double mutation made on a GEP 1 background (GEP 1 m ) did not affect T m , only marginally increased affinity by about three-fold, and did not improve neutralization potency (Figs. 2A, 2D, 6 and Table 1), arguing that affinity and neutralization are largely influenced by somatic mutations through indirect effects beyond the ability to adopt the optimal binding conformer.

GEP structures revealed that structural plasticity was retained during affinity maturation
In order to determine whether the 4E10 combining site rigidified during affinity maturation, undergoing binding site preconfiguration, the crystal structures of GEP 7 and GEP 1 were determined in the unbound state at resolution values of 1.9 Å and 1.7 Å respectively, rebuilt and refined with good statistics ( Table 2), and compared to the unbound structure of 4E10 (4LLV.pdb [17]). Superpositions of 4E10 and GEP bound and unbound structures showed that interdomain movements, while present, were limited compared to CDR rearrangements ( Fig. 9), and that 4E10 retained at least as much CDR flexibility as was present in GEPs, particularly in HCDR2 and HCDR3, while LCDRs were relatively constrained (Figs. 10A and 10B). Calculated root mean square deviations (RMSDs; Fig. 10C) of V H and V L superpositions, with and without CDRs, confirmed that while the FWR regions of GEPs and 4E10 were nearly identical, the bulk of rearrangements observed during binding occurred in HCDRs, with 4E10 movements as large, or larger, than observed in GEPs. LCDR movements, while much smaller overall, contributed less to rearrangements in 4E10 than GEPs, suggesting some minimal degree of rigidification during maturation. Comparison of the surfaces directly contacting epitopes (Movies S1 and S2) showed that GEPs retained higher degrees of structural conservation than 4E10 between the bound and unbound states. These analyses need to be interpreted cognizant of the constraints imposed by crystallization, which involved  Table 1 varying non-physiological solution conditions and variable crystal contacts between structures. However, these caveats were minimized by comparing multiple structures with multiple copies per asymmetric unit.

Engineered ESs can make extensive contacts outside of the targeted epitope
The goal of reverse engineering an Ab is to scaffold the desired epitope to re-elicit Abs that solely recognize the epitope [6,57]. However, the 4E10 linear epitope, as currently defined, is smaller than typical Ab/antigen interfaces, which poses the design challenge of isolating humoral responses to the epitope and not contiguous scaffold surfaces. Previous crystal structures of 4E10/ ES complexes showed that many ESs achieved this goal well, including T93 [15,55]. However, GEP/T117 complex structures reported here showed extensive GEP/scaffold contacts (SA for GEP2 contacts to scaffold minus epitope = 308 Å 2 ) (Fig. 11). A dominant feature of these interactions was the binding of the sidechain of the GEP-specific residue F54H in a deep hydrophobic cleft of the T117 scaffold protein, a putative phosphotransferase from S. typhimurium. These additional contacts raise the concern that Abs elicited by T117 immunizations may have off-target (non-Env) specificities. Nevertheless, the T117 scaffold is highly complementary to 4E10 (Sc for GEP2/T117 = 0.69), which may help explain the increased affinity of T117 for 4E10 and GEPs, and may be ideal for preferentially targeting GEPs through the F54H interaction (F54H is present in the heavy chain gene used by all GEPs). The additional T117 contacts appear to have had the effect of increasing the affinity for T117 over other ESs by two orders of magnitude for 4E10 and by one to three orders of magnitude for the GEPs. However, structural superpositions show that these extra Ab/scaffold interactions do not affect Ab/epitope interactions, or the structure of the epitope in the complexes, which are highly conserved (Fig. 8B).
GEPs displayed a similar degree of limited polyspecificity to 4E10, but different autoreactivity A validated phage-displayed library consisting of 413,611 overlapping 36-mer peptides spanning the entire human proteome combined with phage immunoprecipitation sequencing (PhIP-Seq) was used to assess the polyspecificity and autoantigen recognition profiles of GEPs in comparison to 4E10 [17,50] (Fig. 12A, Tables 5, 6, and 7). GEP 2 and GEP 4 were selected to represent both the clustering (GEP 2) and 96H mutant (GEP 4) GEPs, and an affinity-matured, murine anti-canine CD28 Ab, 1C6 [58], was included for comparison (Fig. 12A, Table 8). The top hit in the 4E10 Fv PhIP-Seq analysis reported here, a peptide derived from the type 2 inositol 1,4,5-trisphosphate receptor (IP 3 R), matches the top hit from the previous PhIP-Seq analysis of IgG 4E10 [17]. However, overall scores were considerably reduced in the PhIP-Seq analysis of 4E10 as an Fv construct versus intact IgG (replicate average 2Log 10 P-values for the top ten scoring peptides of 35.3  Table 1 to 255 for IgG versus 4.35 to 12.3 for Fv), likely due to decreased accessibility of coupled Fv relative to IgG, an increased chance of inactivating Fv versus IgG during chemical coupling, and the inherent increase in local avidity of bivalent IgG versus univalent Fv on potentially sparsely-coupled beads. Given these caveats, overall scoring behavior was very similar across the Fvs tested, with 4E10 and 1C6, the affinity-matured Abs, showing the highest average scores. GEP 4 and 1C6 showed the largest number of high-scoring hits, with 61 and 194 peptides with replicateaveraged 2Log 10 P-values of $4.0 respectively; 4E10 and GEP2 had 12 and 20 peptides, respectively, scoring $4.0. Qualitatively, the results were not dramatically different, but with GEP 4 and 1C6 showing nominally greater spreads of top-scoring peptides rising above the bulk responses. None of the top-scoring three-dozen 4E10 peptides appeared in the top three-dozen hits from either GEP; however two of the top-scoring dozen GEP peptides (derived from zinc finger Ran-binding domain-containing protein 3 or hyaluronidase-3 isoform 1 precursor) were in common between GEP 2 and GEP 4, scoring 1 and 12 (GEP 2) and 4 and 2 (GEP 4), respectively. However, no common peptide motifs could be identified within or between Fv results. None of the IP 3 R peptides scored in the top 65,000 GEP hits. While 1C6 showed the highest scoring spread of top hits (replicate average 2Log 10 Pvalues for the top ten scoring peptides of 12.14 to 30.9), the top ten scoring peptides displayed a considerably higher average hydrophobic character (W), with average W values of: 4E10 = 0.37; GEP 2 = 0.42; GEP 4 = 0.45; 1C6 = 0.71 (higher values are more hydrophobic). Using relative hydrophobicity of the top-scoring PhIP-Seq peptides as a surrogate measure of the overall hydrophobicity of the combining site was consistent with the structures of the Abs (Fig. 12B), where the 1C6 combining site is structured as a large, broad, very hydrophobic concavity and 4E10 and its GEPs sporting smaller, convex hydrophobic surfaces.
Pre-binding of 4E10 at the MPER affects the binding of bnAb b12 at the CD4 binding site In order to test the hypothesis that 4E10 may induce global conformational changes in gp140 3 as part of a higher-order neutralization mechanism [17,21], consistent with downstream effects such as gp120 release, binding of Abs to epitopes distant from the MPER (the V3 loop, by 447-52D [59], and the CD4 binding site, by b12 [60]) was assayed by SPR in the presence or absence of saturating 4E10 (Fig. 13). The qualitative results show that 4E10 pre-binding does not affect 447-52D binding to the flexible, extended V3 loop, but does alter the association kinetics of b12 at the CD4 binding site, suggesting that 4E10 binding induced global conformational changes in gp140 3 registering at distant sites. Since b12 dissociation kinetics were unaffected, 4E10 did not preclude achieving a similar b12 bound-state conformation.

Discussion
Reverse engineered ESs are ideal reagents for studying the binding properties of GEPs. Where peptides and Env do not  Table 1   display sufficient affinities, the best ESs display strong affinities across the GEP ensemble, allowing for the biophysical characterization of 4E10 ontogeny by using ESs to compare 4E10 and GEP binding kinetics and bound-state crystal structures. Consistent with current theory and previous results comparing GEPs with matured bnAbs or other Abs, 4E10 displays orders-of-magnitude better affinities than candidate GEPs for both Env gp140 3 proteins and engineered ESs. This gain in affinity is potentially sufficient to account for the concurrent gain in neutralization potency over GEPs. Previous results showing minimal contributions from FWR mutations [61] and comparisons of complex structures of 4E10 with GEP/T117 complex structures, which showed near-complete conservation of the recognition interface (Fig. 8), argue that affinity maturation was likely due, at least in part, to the small number of somatic mutations in the CDR regions, e.g. 32L and 95H, that add or improve direct contacts to epitope. This limited number of mutations is presumably readily achievable during conventional vaccination. Based on PhIP-Seq peptidome binding results (Fig. 12A), 4E10 and GEPs have distinct autoantigen profiles, suggesting that 4E10 autoreactivity was acquired during ontogeny and not inherent in GEPs, consistent with recognizable GEPs populating naïve repertoires of mature B cells at relatively high frequencies [16]. Among ESs, T117 in particular interacts with GEPs sufficiently strongly to drive B cell activation and maturation (Fig. 2D) and selectively interacts with GEP-specific structural features, e.g. F54H (Fig. 11). These aspects of 4E10 combine to seemingly argue that 4E10 ontogeny follows a relatively short path and that directed re-elicitation of 4E10, and perhaps other bnAbs, may be an achievable goal.
In many aspects, however, 4E10 ontogeny appears more convoluted. Contrasting with the consensus model of Ab ontogeny and results from other systems [38], 4E10 is considerably less thermostable than GEPs, which was difficult to account for only from static views of structures. This lowers the headroom available to tolerate mutations that increase affinity or potency that otherwise might degrade stability, potentially adding a significant hurdle to re-elicitation of 4E10, but clearly does not limit the ability of 4E10 to bind to or neutralize HIV. The conformer ensemble sampled by GEP combining sites echoes the startling plasticity of 4E10, with structural comparisons of bound versus unbound states (Figs. 9 and 10) showing perhaps a small degree of 4E10 LCDR rigidification, but an increase in the conformer ensemble sampled by the HCDRs. This directly contradicts the consensus model, and specific examples [40], of Ab maturation, showing alternatively that preconfiguration of 4E10 does not occur during, or contribute to, affinity maturation. HCDR flexibility is highlighted by F29H, which is able to flip out into solvent in the unbound 4E10 and GEP 1 structures, allowing HCDR1 to dynamically sample multiple conformers. A high number of conserved glycine residues may contribute to 4E10 and GEP HCDR mobility. The reduced affinities due to decreased k a s (Fig. 2D), relative to the clustering GEPs, of GEP 4 and 6 with non-glycine residues at position 96H, predicted to restrict HCDR3 conformer sampling, suggests that HCDR3 mobility was needed to achieve the bound-state conformation with optimal kinetics by destabilizing non-optimal conformers. Retention of significant combining site plasticity also strongly argues for a functional role other than polyspecificity, which is unremarkable in comparison to other Abs, based on PhIP-Seq peptidome binding results (Fig. 12A). Minimal focusing of apparent polyspecificity also contradicts the consensus model, and specific examples [62], of maturation. Also unexpectedly, the ,100-fold increase in affinity of 4E10 over GEPs to T117 is through a decreased k d (Fig. 2D), suggesting that improvements in binding affinity arose from a more favorable bound state and not a decrease in entropic barriers imposed by more plastic GEPs. Thermodynamic studies of 4E10 and GEPs were not possible because no single antigen, needed for valid comparisons, bound to both 4E10 and GEPs with parameters accessible to measurement. However, while crystal structures showed a highly-conserved bound state in both 4E10 and GEPs, the small number of improved contacts could not account fully for the observed improvement in affinity. Complicating this analysis, and its use as a vaccine immunogen, are the extensive contacts between GEPs, and presumably also 4E10, and the scaffold of T117 outside of the stabilized MPER epitope (Fig. 11). The interaction with T117 also highlights another possibility. The 4E10 combining site extends beyond the minimal, linear epitope in T117, making contacts to the scaffold. Unless the MPER epitope forms an isolated structure extended away from the rest of Env, 4E10 may make contacts to elements of Env outside of the linear 4E10 epitope, as it does with T117. HCDR conformer dynamics may therefore be understood as enabling  such interactions, which may foster conformational changes in Env leading to neutralization by any of several mechanisms. Prebinding of 4E10 to gp140 3 affected b12 binding at the CD4 site distant from the MPER (Fig. 13), indirectly demonstrating a global conformational change consistent with this supposition. Additional interactions between GEPs and the scaffold moiety of T117 would also be predicted to drive rigidification of an even greater portion of the combining site during conventional immunization, increasing the distinction between 4E10 and a matured Ab derived from a 4E10 GEP stimulated with T117. T117 was the product of a cutting-edge design effort to generate a structurally-stabilized MPER epitope with optimized binding properties for use as a vaccine immunogen to drive 4E10 ontogeny based on the current paradigm of affinity maturation. However, 4E10 appears to take the affinity maturation pathway less traveled, one that contrasts, in many fundamental ways, the current paradigm. This raises the concern that conventional immunization protocols, based on the current paradigm of structural stabilization of optimal binding conformers, will not successfully re-elicit 4E10 or other unconventional Abs. Understanding unconventional maturation pathways then becomes paramount for the future of molecular vaccinology, allowing efforts to focus on re-eliciting Abs identified as the products of conventional ontogeny in the near term while developing unconventional vaccination strategies to target exceptional Abs in the long term.

SPR interaction analyses
All SPR experiments were performed at 25uC on a Biacore T100 instrument with the T200 sensitivity enhancement (GE Healthcare). For analyses of the binding of 4E10 and GEP Fv analytes to chip-captured ES ligands (Fig. 2D, Table 1), ESs at The interaction partners were separated as indicated in (A) to reveal details of the interface. T117, on the left, is shown in a red semi-transparent molecular surface representation with the NWFDIT epitope, in a licoricestick representation, visible within the surface. The GEP 2 Fv structure is shown on the right, colored as in (A). The NWFDIT epitope, excised from the T117 structure, is shown as bound for reference. The footprint of GEP 2 (non-epitope contacts) on T117 is colored dark grey; the footprint of T117 to non-epitope residues on GEP 2 is colored light grey. The position of the side-chain of F54H is indicated. doi:10.1371/journal.ppat.1004403.g011 ,1 mg/mL were captured on a SA sensor chip (GE Healthcare) following either carboxy (T117) or amine (T72, T93, T344) biotinylation, following the manufacturer's recommended protocol (EZ-Link, Thermo Scientific), in a running buffer of HBS-EP+ (10 mM HEPES, pH 7.4, 150 mM NaCl, 3 mM EDTA, 0.05% P-20; GE Healthcare) plus 0.1 mg/mL bovine serum albumin. A reference flow cell was left blank. Duplicate 4E10 and GEP Fv analyte injections were randomized and run at a flow rate of 50 mL/minute. Regeneration, if needed, was achieved by injection of 10 mM glycine at 50 mL/minute followed by buffer stabilization. Sensorgrams obtained from SPR measurements were doublereference subtracted [69] with BIAevaluation 2.0.3 software (GE Healthcare) employing previously described methodology [18]; data were fit with either 1:1 or steady-state binding models. For analyses of the binding of 4E10 and GEP Fv analytes to gp140 3 (Fig. 2C), gp140 3 at 30 mg/mL in 10 mM sodium acetate (pH = 5.0) were direct amine coupled at a density of ,2200 RUs on a CM5 sensor chip (GE Healthcare); a reference surface was prepared by activating and deactivating a flow cell without the addition of protein; gp140 3 were prepared as previously described [18]. Duplicate 5 minute, 300 nM injections of 4E10 and GEP Fv were made in HBS-EP+ at 50 mL/minute followed by 5 minutes of dissociation, regeneration with 10 mM glycine (pH = 1.5) for 5 seconds and then 6 minutes of stabilization. Assays of Fab b12 or Fab 447-52D binding to SF162 gp140 3 in the presence or absence of 4E10 Fv (Fig. 13) were conducted in HBS-EP+ buffer. Using standard amine coupling chemistry, 1273 RUs of SF162 gp140 3 were coupled to a CM5 sensor chip. A reference surface was prepared by activating and deactivating a flow cell without the addition of protein. Two samples were injected at a flow rate of 50 mL/minute using the dual injection command in the T100 control software (v2.0.3, GE Healthcare) with injection 1 at 5 minutes, injection 2 at 2 minutes and a final dissociation time of 5 minutes. Fab-alone curves were generated by injecting HBS-EP+ followed by an injection of 10 nM Fab b12 or 447-52D and double referenced by subtracting a dual injection of HBS-EP+ followed by HBS-EP+. Fab with 4E10 Fv curves were generated by injecting 3 mM 4E10 Fv (,90 RUs bound) followed by an  [80]; patches are colored in descending order of total area (red, orange, yellow, …). The total surface area for red and orange hydrophobic patches is shown. The crystal structure of GEP 7 is partially disordered in HCDR1 and 3 and so patch area is underrepresented in the calculation. doi:10.1371/journal.ppat.1004403.g012 injection of 10 nM Fab b12 or 447-52D in the presence of 3 mM 4E10 Fv and double referenced by subtracting a dual injection of 3 mM 4E10 Fv followed by 3 mM 4E10 Fv. Optimal regeneration was achieved by injection of 10 mM glycine (pH = 1.5) at a flow rate of 50 mL/minute for 5 seconds followed by a 6 minute buffer stabilization phase. Sensorgrams were corrected by the doublesubtraction method [69] in Scrubber 2.0b software (BioLogic Software). Fig. 13 shows the second sample of each set of injections with baselines zeroed just prior to the second injection.  Crystals were cryopreserved in mother liquor containing 15% v/v glycerol (GEP 1, GEP 1/T117, GEP 7/T117), mother liquor containing 10% (1C6) or 20% (GEP 7) v/v glycerol, or 20% w/w sucrose (GEP 2/T117). Diffraction data for GEP 1, 4E10, and 1C6 were collected at the Advanced Light Source beamline 5.0.2 (Lawrence Berkeley National Laboratory, Berkeley, CA) and reduced using HKL-2000 [70] or d*TREK (Rigaku) [71]. Diffraction data for GEP 7, GEP 1/T117, GEP 7/T117 and GEP 2/T117 were collected in house with CuKa radiation on a R-AXIS IV++ image plate detector with HR optics (Rigaku) at 2170uC. Initial phase information for all data sets was determined by molecular replacement, as implemented in the CCP4i program suite [72][73][74], using 3LF6.pdb (T117), 3LH2.pdb (GEPs) and 1JP5.pbd (1C6) as initial search models. Phases were improved by subsequent rounds of model building and refinement using COOT [75] and REFMAC [76]. Structure validation was carried out with PROCHECK [77], the MolProbity server [78], and the RCSB ADIT validation server. Data collection and structure refinement statistics are shown in Table 2. Crystals of GEP 2 alone could not be grown, despite considerable effort. PhIP-Seq analysis 4E10, GEP and 1C6 Fvs were coupled to beads and analyzed in duplicate. For each Fv analyzed, 3 mg of magnetic beads (Invitrogen M-270 Epoxy Dynabeads) were resuspended in 60 mL 0.1 M NaPO 4 (pH = 7.4). Beads were rocked at ambient temperature for 24 hrs with 60 mg of each Fv in 1 M (NH 4 ) 2 SO 4 and then washed with 10 mM glycine in PBS to cap unreacted epoxy groups. Activity of Fv coupled beads was confirmed by epitope-scaffold binding prior to PhIP-Seq analyses. PhIP-Seq analyses were performed as previously described [17,50]. Results were plotted as replicate #1 versus replicate #2 2Log 10 P-values; highly discordant, and therefore spurious, hits falling near the axes were excluded from analysis. 241 peptides were also discarded because they displayed nonspecific binding to all Abs tested [17]. Peptide hydrophobicity was determined with the Sigma-Aldrich PEPscreen Library Design Tool and overall charge at neutral pH was determined with the Innovagen Peptide Property Calculator.

Supporting Information
Movie S1 Superposition of epitope binding site contacts from bound and unbound forms of 4E10. Residues from epitope binding site contacts are represented as mesh, pink for bound and yellow for unbound. The heavy chain CDR1 and CDR3 loops are shown in cartoon representations with residues W100H, W100bH, and F29H shown in licorice stick representations. (M4V) Movie S2 Superposition of epitope binding site contacts from bound and unbound forms of GEP 7. Residues from epitope binding site contacts are represented as mesh, pink for bound and yellow for unbound. The heavy chain CDR1 and CDR3 loops are shown in cartoon representations with residues W100H, W100bH, and F29H shown in licorice stick representations. (M4V) Figure 13. Effects of 4E10/gp140 3 pre-binding on 447-52D and b12 binding. Double-corrected sensorgrams of the SPR response of 10 nM analytes of Fab b12 (red and green traces) or Fab 447-52D (blue and black traces) to chip-coupled SF162 gp140 3 in the absence (red and black traces) or presence (green and blue traces) of pre-bound Fv 4E10 are shown. Fv 4E10, at a concentration of 3 mM, was flowed over the chip starting 5 minutes before the 447 or b12 injections and continued throughout the experiment. The black and blue 447-52D traces were completely superimposed, showing no effect of 4E10 pre-binding on 447-52D binding. While the dissociation phases of the b12 traces are nearly parallel, showing no minimal effect of 4E10 pre-binding on the dissociation of b12 from the bound state, the association phases are clearly not parallel, resulting in a lower peak response in the presence of 4E10 pre-binding, showing a modest, but important, reduction in the association rate. This result suggests that the CD4 binding site adopted a distinct conformational state prior to b12 binding as a consequence of 4E10 binding at the MPER. doi:10.1371/journal.ppat.1004403.g013