Predictive measures of hydrophobicity, as determined by scales of free energy for partitioning whole residues from water to a lipid bilayer interface, ΔGwif
, or to octanol, ΔGoct
), were used to design mutations that would either disrupt or augment the predicted partitioning of the 2F5 CDR H3 tip (residues L100A
) (Fig. ). The amino acid serine, which is found midway on the ΔGwif
hydrophobicity scales, was chosen as a single or double substitution in order to achieve a gradual decrease in hydrophobicity (Fig. and Table ). Conversely, tryptophan, which is the residue most favored to partition into the bilayer interface or octanol (40
), was used to augment the hydrophobicity of the loop (Fig. and Table ).
2F5 CDR H3 partition free energies, binding constants to gp41 peptide, and neutralization IC50se
To ascertain the effect of the CDR H3 tip mutations on binding of 2F5 to the gp41 MPER, the affinities of the expressed 2F5 variants for a gp41 MPER peptide were determined using surface plasmon resonance. The 2F5 wild-type and mutant IgGs were coupled directly to the biosensor surface, and a gp41 MPER peptide comprised of residues 657 to 669 of gp41 (isolate HxB2 numbering) linked to a C-terminal C9 tag was used as analyte, at 2-fold serial dilutions ranging from 500 to 0.49 nM. As shown in Table (see Fig. S1 in the supplemental material), all of the mutants maintained nanomolar affinity to the gp41 peptide. In some cases, such as for mutants of I100F, there was a moderate reduction in affinity, likely due to minor contacts made by this residue with gp41.
The 2F5 mutants were then tested for neutralization of the laboratory-adapted strain HxB2, which is highly sensitive to wild-type 2F5. A single-cycle infectivity assay using Env-pseudotyped virus and Tzm-bl target cells was employed for this purpose, as previously described (18
). As shown in Fig. , top, even single replacements of hydrophobic residues of the CDR H3 loop with serine were able to reduce the neutralization capacity of 2F5 against HxB2 by several orders of magnitude. The most noticeable single-mutation effect was observed for the F100B
S mutation, which led to a more than 500-fold increase in the 50% inhibitory concentration (IC50
) relative to wild-type 2F5 (Fig. , top, and Table ). The single mutations L100A
S and I100F
S likewise led to 100-fold increases in the IC50
s, while the V100D
S mutation increased the IC50
by about 15-fold. Introduction of double mutations to serine at the same residue locations, furthermore, led to complete abrogation of 2F5-mediated neutralization of HxB2, with neutralization profiles indistinguishable from those of the negative control mouse 1D4 anti-rhodopsin antibody (20
) (Fig. , middle, and Table ). Thus, modification of residues at the tip of the CDR H3 loop can reduce and completely disrupt 2F5 neutralization, despite maintaining nanomolar affinity for gp41.
FIG. 2. Neutralization of HxB2 by CDR H3 mutants of antibody 2F5. (Top) Neutralization profiles of 2F5 variants with single mutations to serine. Single serine substitutions resulted in a 15- to 500-fold reduction in neutralization potency. (Middle) Neutralization (more ...)
Since these findings could be the result either of a disruption of specific protein contacts made by these residues or of a disruption of nonspecific interactions mediated by their chemical nature, mutation of the same CDR H3 residues to tryptophan was undertaken. Substitutions to tryptophan served two purposes: first, they augmented the hydrophobicity and predicted favorability of free energies of transfer from water to a bilayer interface or to octanol (ΔGwif and ΔGoct), and second, the bulky nature of the tryptophan side chain had the potential to disrupt protein-protein contacts, should they exist. When tested for neutralization of HxB2, the tryptophan mutants were either commensurate with 2F5 wild-type neutralization, such as L100AW and F100BW, or were even more potent, such as V100DW and the double mutant L100AW V100DW, both of which showed an ~10-fold decrease in the neutralization IC50 relative to wild-type 2F5 (Fig. , bottom).
To rule out the possibility that these results were specific to the HxB2 isolate, the 2F5 mutants were tested for neutralization of a panel of tier 1 and tier 2 HIV-1 isolates, ranging from highly sensitive strains, such as MN, to more resistant ones, such as JR-FL (17
), and also to an HIV-2-HIV-1 2F5 epitope chimera, 7312a-C1 (6
). As shown in Table , single mutations that decreased the hydrophobicity of the 2F5 CDR H3 loop led to decreases in neutralization potency, and double mutations that decreased hydrophobicity completely abrogated neutralization. Likewise, single and double mutations that increased the hydrophobicity of the loop led to increased neutralization potencies (Table ). Meanwhile, neutralization of the Tro.11 isolate, which has a K665S point mutation in the core of the epitope, was virtually undetectable for all 2F5 mutants, similar to what was observed for viruses pseudotyped with MuLV and parental HIV-2 7312a Env, which were used as negative controls.
To determine if there was a statistical relationship between the hydrophobicity of the 2F5 CDR H3 loop and the neutralization capacity of the antibody, neutralization IC50s of the 2F5 mutants were plotted against the estimated free energies of transfer, ΔGwif, of the mutant CDR H3s for each of the strains tested. As shown in Fig. and Table , with the exception of RHPA-4259, statistically significant linear relationships were observed between neutralization IC50s and CDR H3 ΔGwif, with P values ranging from 0.0012 for HxB2 to 0.018 for JR-FL. Linear relationships with even more stringent P values were observed between neutralization IC50s and the predicted free energy of transfer to octanol, ΔGoct (see Fig. S2A and Table S1 in the supplemental material). Associations between the neutralization IC50s and the affinities of the various 2F5 mutants for gp41 MPER peptide were also examined. As shown in Fig. and Table , despite the fact that the fits appear to be largely driven by the KD (dissociation constant) of the I100FS 2F5 mutant for gp41 (which makes minor contacts with the peptide), no statistically significant relationships were observed between the neutralization IC50s of the 2F5 CDR H3 variants and their affinities for the gp41 MPER peptide.
FIG. 3. Relationship between 2F5-mediated HIV-1 neutralization and the hydrophobicity of its CDR H3 loop. (A) 2F5 variant neutralization IC50s plotted against the calculated ΔGwif of the 2F5 CDR H3 loop for each virus strain tested. Linear regressions (more ...)
Linear correlation statistics of neutralization IC50 and ΔΔGN versus ΔGwif and KD gp41 peptidea
Though the overall levels of the IC50s varied per strain tested, likely a reflection of the strain sensitivity to 2F5 itself, the mutations appeared to exhibit similar effects on neutralization regardless of the strain used. Specifically, no significant differences were observed in the slopes of the linear fits of the neutralization IC50s versus ΔGwif or versus ΔGoct, unlike their y intercepts, which did display significant differences (Fig. and Table ; see Fig. S2A and Table S1 in the supplemental material). The similarities in the slopes of the regressions across all strains suggested that the neutralization effects mediated by 2F5 CDR H3 loop hydrophobicity were largely independent of strain sensitivity to 2F5.
The free energy of antibody neutralization (ΔGN
) can be viewed as a sum of free energies that contribute to its functional interactions with a specific strain of virus (13
). It can also be viewed as a binding association of an antibody “n
” with the virion: ΔGN
= −RT ln K
), where K
) is defined as IC50
) and f
) is a function that accounts for variables such as strain sensitivity and assay used. Although an absolute value for this free energy requires the definition of f
), if f
) can be approximated to equal f
) for two variants n1
of the same antibody neutralizing the same strain of virus, a relative free energy, ΔΔGN
, can be obtained (13
). In the case of 2F5, we sought to verify how the free energy of partitioning the CDR H3 loop from water to a lipid bilayer interface, ΔGwif
, or to octanol, ΔGoct
, correlated with the relative free energy of neutralization (ΔΔGN
). Linear models were used to fit curves of ΔΔGN
(Fig. ) or ΔGoct
(see Fig. S2C in the supplemental material) for each individual strain and for all strains together. The shared correlations obtained were statistically significant, with P
< 0.0001, although end points at high hydrophobicity suggested linear fits might not be ideal (see below). Nonetheless, the results confirmed that the effects of ΔGwif
of the 2F5 CDR H3 loop on virus neutralization were largely independent of the virus strain, with the normalization relative to 2F5 wild type in the ΔΔGN
calculation making the fits for all the strains virtually superimposable (Fig. ; see Fig. S2C in the supplemental material). The shared slopes from these correlations were 1.3 and 1.2 for ΔGwif
, respectively. Thus, the calculated change in the free energy of partitioning the 2F5 CDR H3 loop translates almost directly into changes in neutralization. The 30% enhancement in neutralization, ΔΔGN
, over the calculated partition free energy may reflect the planar positioning of these residues in the CDR H3 structure and/or their positioning induced by recognition of the gp41 protein component (42
Though linear regressions provided a reasonable first approximation of the relationship between neutralization IC50
s and the free energy of hydrophobic transfer of the tip of the 2F5 CDR H3 loop, we observed that beyond a certain threshold of loop hydrophobicity, the effects on neutralization appeared to level off. To account for this observation, a quadratic term was added to the linear regressions. This yielded better fits of the data, as judged by an extra sums-of-squares F test (Fig. ; see Fig. S2B and Table S2 in the supplemental material) (15
). Based on these quadratic models, IC50
minima were interpolated for each of the strains tested (see Table S3 in the supplemental material). On average, the interpolated minimum IC50
s were approximately 0.96 log units or 9.2-fold lower than the corresponding experimental IC50
s of wild-type 2F5 (see Table S3 in the supplemental material). Compared to the interpolated IC50
s corresponding to a ΔGwif
of 0, in which no transfer is predicted to occur, the mean minimum IC50
s were approximately 5.1 log units or 13,000-fold lower than those predicted for a 2F5 variant with no capacity for hydrophobic transfer (see Table S3 in the supplemental material).
FIG. 4. Threshold for effects of increased 2F5 CDR H3 loop hydrophobicity. To account for the observed leveling off of the effects of increased 2F5 CDR H3 loop hydrophobicity, quadratic fits were applied to the data shown in Fig. . (A) 2F5 variant (more ...)
A quadratic term was also added to the fits of the plots of the relative free energies of neutralization, ΔΔGN, versus the predicted free energies of transfer of the 2F5 CDR H3 loop to a bilayer interface, ΔGwif, or octanol, ΔGoct (Fig. ; see Fig. S2D in the supplemental material). Shared quadratic fits of ΔΔGN versus ΔGwif and ΔGoct were also performed, and ΔΔGN minima were observed at 2F5 CDR H3 ΔGwif and ΔGoct values of −4.08 and −5.69 kcal/mol, with corresponding ΔΔGN values of −1.28 and −1.18 kcal/mol, respectively (Fig. ; see Fig. S2D in the supplemental material).