Discovery at the VICB
Better HIV Antibodies by Design
By: Carol A. Rouzer, VICB Communications
Published: June 1, 2015
Computational modeling predicts mutations that increase the affinity and broaden the specificity of anti-HIV neutralizing antibodies.
A major challenge in the fight against AIDS (acquired immunodeficiency syndrome) is the ability of the causative agent HIV (the human immunodeficiency virus) to evade the immune system. One mechanism that the virus uses to accomplish this goal is the resistance of its Env protein to antibody attack. Env is a surface glycoprotein comprising a trimer of the transmembrane gp41 subunit capped by a trimer of gp120 (Figure 1). Env plays a critical role in the fusion of HIV with a target cell membrane. Most antibodies that neutralize HIV bind to gp120, suggesting that immune defense against the virus is possible if a vaccine can be designed that leads to the efficient production of gp120-targeted antibodies. However, this goal has been frustrated by the presence of the gp120 V1/V2 domain that has a highly variable sequence, making it difficult to produce single antibodies that are effective against a broad range of viral strains. Now, VICB member Jens Meiler in collaboration with James Crowe of Vanderbilt’s Department of Pathology, Immunology, and Microbiology, have joined to bring a new approach to this vexing problem. [J. R. Willis, et al. (2015) J. Clin. Invest., published online May 18, DOI:10.1172/JCI80693].
Figure 1. Schematic diagram of the human immunodeficiency virus. The viral RNA genome and reverse transcriptase are surrounded by capsid proteins, and the capsid is, in turn, surrounded by the membrane. The Env protein, comprising gp41 transmembrane trimers complexed with gp120 trimers, decorates the outer surface of the virus, forming the viral spike that plays a key role in fusion with a target cell membrane. Reprinted from Wikimedia Commons. Figure produced by the U.S. National Institutes of Health, public domain.
PG9 and PG16 are two monoclonal antibodies that neutralize 70-80% of circulating HIV isolates. Both antibodies target an N-linked glycan at arginine-160 in the V1/V2 region of gp120. All antibodies comprise two heavy chains and two light chains complexed to form a Y-shaped structure (Figure 2). The variable domains of the heavy and light chains at the tips of the Y form the site of antigen binding. Key to the antigen-antibody interaction are the three complementarity-determining regions (CDRs) in the variable domains of each chain that form contact sites for antigen. PG9 and PG16 are notable for their unusually long third heavy chain CDR (HCDR3) that forms a hammerhead-like structure (Figure 3). HCDR3 burrows deeply into the glycan of the V1/V2 domain of gp120, clearly playing an important role in the binding affinity of the two antibodies. Despite this, the HCDR3s of PG9 and PG16 vary by 33%, leading the Meiler and Crowe investigators to hypothesize that additional modifications in this region would be tolerated and might also lead to proteins with higher affinity or activity against a broader range of viral strains.
Figure 2. Diagram of antibody structure. The protein comprises two heavy and two light chains arranged in a Y-shaped complex. Both chains have variable domains at the tips of the arms of the Y where antigen binds. Each variable domain contains three complementarity-determining regions (CDRs), which come into direct contact with the bound antigen. Reproduced from Wikimedia Commons under the Creative Commons Attribution-Share Alike 3.0 Unported license.
Figure 3. Crystal structure of the Fab portions of PG9 bound to the V1/V2 domain of pg120. The heavy and light chains of PG9 are in yellow and blue, respectively. Each CDR in the heavy and light chains is labeled. HCDR3 (labeled CDR H3) can be seen projecting into the glycan attached to asparagine-160 (purple) of V1/V2. Figure reproduced by permission from Macmillan Publishers, Ltd. from J. S. McLellan, et al., (2011) Nature, 480, 336. Copyright 2011.
The HCDR3 loop contains 28 amino acids, so 532 different point mutations of the protein are possible, and there are 2028 variants containing two or more altered amino acids. Clearly, it was not feasible to create and test all of these possible variants, so the Meiler and Crowe team decided to use a computational approach to first identify mutations most likely to increase the strength of the antibody-antigen interaction. Starting with the crystal structure of PG9 complexed to the HIV Env protein, the Meiler group used RosettaDesign to search for mutations that would minimize the Rosetta energy function for the overall complex. Their results pointed to five mutant proteins that were predicted to show substantial improvements in antigen binding without detrimental reductions in overall antibody stability. These proteins were N100FY, N100FL, D100LN, PG9-2MUT (containing A96S and Y100QN) and PG9-4MUT (containing A96S, Y100QN, D100LN, and N100FL). Note that the amino acids are designated using the Kabat system, which numbers residues within HCDR3 between arginine-100 and methionine-101 as 100A through 100S.
The Crowe lab conducted binding studies of each mutant protein using gp120 monomers from HIV clades B and C. They found that the N100FL and N100FY mutants exhibited 2.3- to 14-fold stronger binding to gp120 than that of wild-type PG9. In contrast, D100LN exhibited similar affinity to that of wild-type PG9, while the affinity of PG9_4MUT was reduced by 2- to 100-fold, and PG9_2MUT was essentially inactive. When the various proteins were tested for their ability to neutralize HIV, the N100FY mutant demonstrated increased potency compared to that of the wild-type antibody, and it recognized HIV variants that the wild-type PG9 did not. Of particular interest was the finding that N100FY neutralized 7 out of 9 viruses that lacked the arginine-160-linked glycan, suggesting that the mutant antibody was no longer dependent on this epitope for binding.
A crystal structure of the Fab region of N100FY demonstrated that it adopted the conformation predicted by the computational model and that it retained the HCDR3 hammerhead structure that is found in wild-type PG9. The hammerhead structure appeared to be more ordered than those observed on previous crystal structures of the wild-type protein, suggesting that the mutation increases the stability of the antibody. This hypothesis was supported by the presence of π-π stacking between the mutated tyrosine-100F and residues proline-99 and tyrosine-100A. The hypothesis was further supported by the results of Rosetta modeling, which showed that stabilization was a key factor contributing to the increased binding affinity of the mutant, and differential scanning calorimetry, which demonstrated that more thermal energy is needed to denature the N100FY mutant than wild-type PG9.
The Crowe and Meiler investigators were concerned that the apparent increased binding affinity and broader specificity of the N100FY mutant might be due simply to greater nonspecific binding resulting from the observed higher structural stability of the protein. If that were the case, then N100FY would likely show increased binding to human proteins than wild-type PG9, leading to a high risk that it could trigger an autoimmune reaction. To test this hypothesis, the researchers evaluated the binding of N100FY and wild-type PG9 to human HEp-2 cells in culture. The mutant antibody exhibited no higher tendency to bind to the cells than wild-type PG9, and both antibodies bound to the cells at substantially lower levels than control antibodies known to cause autoimmune reactions. These results suggest that the high affinity and broad specificity of N100FY are the result of improved specific binding.
The Meiler and Crowe lab findings are exciting not only because they have led to the discovery of a more potent antibody against HIV, but also because they reveal the power of computational modeling in the design of antibodies with improved properties. Out of five mutants tested, two exhibited the desired properties of increased binding affinity, suggesting a high (40%) success rate for the modeling approach. This is clearly much higher than the success rate expected for random mutation and testing, so this approach leads to substantial and important savings in time and resources. The Meiler and Crowe groups will continue their efforts through a recent $9 million grant that will enable them to apply their methods to the development of better vaccines against influenza virus.