Previous work identified a site of proteolytic cleavage (clipping) at the lysine at position K100m (Kabat numbering) in the CAP256-VRC26.25 CDRH3 5 (Figure 1A,B). We therefore generated a series of mutants in CAP256-VRC26.25 heavy chain in which that position was changed to every possible residue except cysteine. The antibodies were tested against a panel of Env-pseudoviruses to check for loss of neutralization (Figure 1C and S1A). Neutralization of Env-pseudovirus CAP256.209.c2 was unaffected by any of the mutations. Activity of most mutants against Env-pseudoviruses BG505 and DU156 was decreased 5- to 100-fold; the least affected mutants were further tested against MB539, TH976, PVO.04, and AC10.29. Mutation to alanine, methionine, or glutamine (K100mA, K100mM, and K100mQ) caused a small (<5-fold) overall reduction in potency, while the other mutations greatly reduced or abrogated activity. Changing the lysine to an arginine (K100mR) did not alter the neutralization activity. Of note, the CAP256-VRC26.25 appeared to be carried over in pipet tips during dilution, causing an artifact in which the neutralization activity does not titer out. This led us to adopt a neutralization assay protocol in which pipet tips are changed after each antibody dilution step; doing so was found to be critical for accurate assessment of CAP256-VRC26.25 neutralization potency (Figure S2).
To mimic the exposure to proteases released by dead cells in 19-day CHO cell culture, we passed medium from 19 day CAP256-VRC26.25 CHO cell cultures over protein A and retained the flow-through fraction. The wild-type antibody and mutants K100mA, K100mM, K100mQ, and K100mR (produced in 6 day HEK293expi cultures) were incubated in this protease-containing flow-through material at 37oC for 0 or 72 hours. After additional protein A column purification, treated or untreated mAbs were tested by reducing CGE-SDS (GX II) chromatography analysis to monitor heavy chain cleavage (Gollapudi et al, manuscript submitted). The chromatography trace for wild-type CAP256-VRC26.25 included extra peaks identified as the products of heavy-chain cleavage; 6.4% of the antibody mass was assigned to the clipped products (Figure 2). We observed that the K100mR mutant was also subject to cleavage; this was not unexpected, as lysine and arginine are both targets of serine proteases, the presumptive class of proteases acting on the antibodies. In contrast, the K100mA, K100mM, and K100mQ variants were not cleaved (Figure 2).
The cleavage–resistant variant that showed the greatest potency in the initial neutralization screen, K100mA, was further tested on a 208 multiclade Env-pseudovirus panel. When compared to wild type, this mutant had the same overall breadth and median potency (Figures 3A and S1, and Table S1). At the level of individual viruses, we observed that K100mA gained potency for some viruses and lost potency for others, but in a balanced manner such that the total fraction of viruses sensitive to K100mA was the same as to wild type (Figure S1B). We noted a very high correlation between IC50 values for the two antibodies (Spearman’s rho 0.96, p<0.0001) as well as for IC80 values (Spearman’s rho 0.97, p<0.0001). In addition, we used the GPS-TSP algorithm 12 to predict the impact of K100mA on tyrosine sulfation of the CDRH3, as this post-translational modification is critical for full potency of the antibody (Cai et al, manuscript submitted). Predicted sulfation of wild-type and K100mA were nearly identical (Fig S1C), in agreement with the observed retention of neutralization activity. Because the K100mA mutant retained the full breadth and potency of the wild type, while resisting proteolytic cleavage, we selected this variant for further development.
Half-life extending mutation in lead variant
In addition to the K100mA mutation, we added the LS mutation in the Fc portion of the heavy chain, which was previously shown to increase the half-life of parental CAP256-VRC26.25 upon passive administration in non-human primates 13. In humans, the LS mutations in other antibodies have been shown to increase serum half-life in vivo; the mechanism is related to increased antibody binding to the FcRn receptor 14, 15. LS is not expected to affect neutralization potency. To verify this, we assessed neutralization on the 208 multiclade Env-pseudovirus panel for three pairs of bNAbs: VRC01 and VRC01LS, N6 and N6-LS, and PGDM1400 and PGDM1400LS. As expected, only very minor differences were noted between the antibodies with and without LS (Tables 1, S1A-B), with highly significant correlation between values: for each pair of antibodies, both IC50 values and IC80 values correlated with Spearman’s rho of 0.94 or higher and p<0.0001; and the median IC50s were no more than 2-fold different, which is within experimental error for this assay 24. We therefore proceeded to add LS to the Fc portion of the CAP256-VRC26.25.K100mA construct.
The resulting construct, CAP256V2LS, was produced at high purity under GMP conditions. This clinical-grade product was assessed for neutralization on the 208 multiclade Env-pseudovirus panel and was slightly more potent and broad than research lots of either wild type or K100mA (Figures 3 and S3, Tables 1and S1A-B). This improvement may be due to enhanced purity of the clinical-grade product compared to research laboratory stocks, and/or improved tyrosine sulfation in the CHO cell line used for production (Cai et al, manuscript submitted). Overall, clinical-grade CAP256V2LS is 63% broad on 208 viruses, with high potency and 70% breadth across non-B clade viruses, at a cutoff of IC50<50 mg/ml. At the more stringent cutoff of ID80<1 mg/ml suggested by the AMP trial results, breadth on non-clade B was 44% overall and 59% against clade C pseudoviruses in the panel (Figure 3, Tables S1A-B). The potency of CAP256V2LS is greater than other HIV-1 bNAbs that have been in clinical trials (Tables S1A-B); and its breadth, while lower than some of them, is similar to mAb 10-1074. GMP-grade CAP256V2LS was also tested on a panel of 100 clade C viruses, representing the dominant sequences in southern Africa, with 62% breadth at ID80<1 and a median IC50 of 0.002 and ID80 of 0.007 mg/ml (Figure 3, Table S1C-D). The combination with VRC07-523-LS had a predicted breadth of 90% at ID80<1 on both the multiclade and the clade C panels (Tables 1, S1B, S1D).
Structural analysis of mutant
To understand why the K100mA mutation had only minor effects on neutralization breadth and potency, we examined the structure of wild-type CAP256-VRC26.25 alone and in complex with a closed prefusion Env trimer (Figure 4A-B). In the unliganded Fab structure3, K100m formed an electrostatic interaction with D100g of the CDRH3, while in the bound structure16 the D100g residue of the CDRH3 interacted with the positively charged K169 residue of the HIV-1 Env, thereby breaking contact with K100m (Figure 4B). The location and orientation of the wild-type K100m outside of the apex cavity and facing the solvent prevents any meaningful contact with the Env trimer beyond minimal buried surface area of under 10 Å2 on the backbone. The side chain faced away from the Env apex cavity and away from any potential Env interactions. As expected from this observation, modelling of Ala at position 100m onto the cryo-EM structure resulted in no change in the Env binding interactions. (Figure 4C-D).
Pharmacokinetic and autoreactivity analyses
Since CAP256V2LS is of interest for clinical use in HIV prevention, we tested its pharmacokinetic (pK) profile in rhesus macaques and in FcRn mice. The latter are transgenic mice expressing the human FcRn receptor and are commonly used to assess serum half-life of monoclonal antibodies 17. Rhesus macaques were infused with 10 mg/kg of CAP256V2LS and parental CAP256-VRC26.25 with LS (CAP256LS); antibody concentrations above 10 mg/ml were maintained for three weeks post-infusion for both constructs (Figure 5A). This is similar to published values for VRC07-523LS, which is a potent CD4 binding site antibody that is under clinical investigation for both HIV-1 prevention and therapy 18. Concordant data were observed in a second model, human FcRn transgenic mice, in which CAP256V2LS had pK curves similar to VRC07-523LS (Figure 5B). We also tested for autoreactivity in vitro, which can be predictive of poor pharmacokinetic properties in vivo 19 CAP256V2LS was negative in cardiolipin binding and HEp-2 cell staining, indicating a lack of autoreactivity (Fig 5C-D).