In vivo affinity maturation of the CD4 domains of an HIV-1-entry inhibitor

Abstract

Human proteins repurposed as biologics for clinical use have been engineered through in vitro techniques that improve the affinity of the biologics for their ligands. However, the techniques do not select against properties, such as protease sensitivity or self-reactivity, that impair the biologics' clinical efficacy. Here we show that the B-cell receptors of primary murine B cells can be engineered to affinity mature in vivo the human CD4 domains of the HIV-1-entry inhibitor CD4 immunoadhesin (CD4-Ig). Specifically, we introduced genes encoding the CD4 domains 1 and 2 (D1D2) of a half-life-enhanced form of CD4-Ig (CD4-Ig-v0) into the heavy-chain loci of murine B cells and adoptively transferred these cells into wild-type mice. After immunization, the B cells proliferated, class switched, affinity matured and produced D1D2-presenting antibodies. Somatic hypermutations in the D1D2-encoding region of the engrafted cells improved the binding affinity of CD4-Ig-v0 for the HIV-1 envelope glycoprotein and the inhibitor's ability to neutralize a panel of HIV-1 isolates without impairing its pharmacokinetic properties. In vivo affinity maturation of non-antibody protein biologics may guide the development of more effective therapeutics.

Extended Data Fig. 1. Strategies for presenting human biologics on the BCR of primary murine B cells.

a Designs of CD4 D1D2-presenting B-cell receptors. Below each structure is a representation of the cassette used to generate it. D1D2 (yellow) was attached to the N terminus of OKT3 heavy-chain variable region (CD4-OKT-V_H, green, left) via a (G₄S)₃ linker or attached directly to OKT3 kappa light-chain constant region (light blue, center and right). This light-chain constant region was linked to that of the heavy chain via a (G₄S)₃ linker (CD4-Cĸ-GS, center), or non-covalently associated with heavy-chain constant region by P2A cleavage (CD4-Cĸ-P2A, right). b The editing strategy used to express these constructs in primary murine B cells. A gRNA targets an intronic region immediately downstream of the J4 segment. Repair template homology arms complement this region, facilitating insertion of a cassette with a poly-A tail, a splice acceptor (SA), a heavy-chain promoter, a D1D2 construct, and a splice donor (SD). Although this editing strategy is efficient, it interferes with affinity maturation of the expressed B-cell receptor. An alternate strategy, used hereafter, is represented in Fig. 1b. c Flow cytometric analysis of singlet viable B cells edited using the strategy in b were stained with anti-IgM antibodies and either anti-human CD4 antibodies or HIV-1 gp120 at 48 h post-editing. d Expression of multiple biologics on primary mouse B cells. Using the editing strategy presented in Fig. 1b, OKT-V_H fusions of human SIRPα, CTLA-4, and IL-7 were introduced into the heavy-chain locus of primary murine B cells. HDRT were delivered with AAV-DJ. Cells were analyzed 48 h post-electroporation by flow cytometry for their ability to bind human CD47, human CD80, or anti-IL-7 antibodies. Note that CTLA-4 and IL-7 have murine receptors that can bind the secreted antibody fusion to recognize both edited and unedited cells. a, b are created on BioRender.com.

Extended Data Fig. 2. Neutralization of mouse serum and CD4-OKT-IgG to various pseudoviruses and immunization regimen of mice receiving protein immunogens.

a Neutralizing response against HIV-1 CE1176 pseudovirus. Sera collected from mice (M1-M5, n = 5) engrafted with edited cells and immunized twice with mRNA-LNP at two-week interval were tested in TZM-bl neutralization assays and represented as individual curves. Sera from mock mice (n = 3) engrafted with unedited CD45.1 B cells and immunized twice at two-week intervals (grey dots) served as negative controls. Dots and error bars of negative controls indicate median and interquartile range. Positive control CD4-Ig-v0 was mixed with naïve mouse sera at 125 μg/ml. Error bars represent SEM. b Neutralizing potency of CD4-OKT3-Ig. CD4-OKT3-Ig purified by size exclusion chromatography (SEC) was compared with CD4-Ig-v0 for IC₅₀ for BG505 and TRO11 pseudoviruses (PVs). c Engraftment and immunization regimen for mice edited with the same method as Fig. 1 and vaccinated with either mRNA-LNP (M1-M10) or protein antigens (M11-M24). Mice were engrafted with 15,000 (M11-M15) or (M16-M24) successfully edited B cells, as determined by flow cytometry analysis. These mice were vaccinated with protein antigens (monomeric or decametric ConM and TRO11 gp120) at two-week interval. Schedule of blood and tissue collection is the same as in Fig. 2a. Decameric antigens are indicated in blue, and monomeric antigens in yellow. d Sera neutralizing response of mice in c against BG505 PV. Sera from mice (n = 3) engrafted with unedited CD45.1 B cells and immunized three times with decametric ConM gp120 (grey dots) served as mock controls. Dots and error bars indicate median and interquartile range. Note that protein immunization induced neutralizing responses only in mice (M16-M24) engrafted with a higher number (500,000) of cells.

Extended Data Fig. 3.. Analysis of antigen binding B cells after three immunizations.

a Gating strategies for flow cytometry analysis of donor and recipient antigen binding B cells and germinal center B cells from spleen and lymph nodes in recipient mice. The top panel represents the gating for singlet viable Ig+ B cells after B cell enrichment for Fig. 3a–b and Ext Fig. 3b. The bottom panel represents the gating for singlet viable CD19+ CD138_low GC B cells for Fig. 3c–d and Ext Fig. 3c–d. b Flow cytometry analysis of antigen binding B cells in singlet viable IgG+ donor (CD45.1) and recipient (CD45.2) population for Fig. 3a–b. c Flow cytometry analysis of germinal center (GL7+ CD38−) B cells from spleen and lymph nodes harvested four days after three immunizations. M1 and M2 were from 2 wk group; M7 and M9 were from 4 wk group. d Flow cytometry analysis of antigen binding GC B cells in M7 and M9, as quantified in Fig. 3d. M1 and M2 were shown in Fig. 3c.

Extended Data Fig. 4.

a Analysis of AID motif and convergence of the top nucleotide mutations among 10 mRNA-immunized mice. The first two columns show the original amino acid and nucleotide. AID hotspot motifs are defined as DGYW / WRCH (R=A/G, Y=C/T, and W=A/T). Mutation frequency is calculated as an average from 10 mice immunized by mRNA-LNP every two or four weeks. Y/N indicates whether the mutation is located at AID motifs. Note that cells obtained from M6 and M7 were combined before sequencing analysis and represented here as a single mouse. Non-synonymous (Non-syn) mutations are labeled in light blue; synonymous (Syn) mutations are labeled in light orange. b Frequency of transitions and transversions in D1D2. Original nucleotides are at the left column, and the hypermutated nucleotides are at the top row. The frequency is derived from synonymous mutations, normalized to 100% for each original nucleotide, and averaged across all mice (M1-M10, M16-M24).

Extended Data Fig. 5.. Diverse and convergent amino-acid mutations in engrafted mice.

a Eight residues with the highest mutation rate across D1D2 from mice (M3 through M10) that were not presented in Fig. 5b, immunized with mRNA-LNP at two-week and four-week intervals. Fewer B cells were isolated from M6 and M7, and these cells were therefore combined for one sample for sequencing. b Eight residues with the highest mutation rate from the rest of mice (M16 through M24) immunized with adjuvanted proteins at two-week and four-week intervals, as shown in Extended Data Figure 2c. Note that mice M11-M15 were engrafted with low numbers of successfully edited B cells (15,000) like M3-M10; M16-M24, engrafted with 500,000 edited B cells, were immunized with protein antigens. No gp120-binding donor cells were isolated from mice M11-M15, and they were therefore excluded from NGS analysis.

Extended Data Fig. 6.. Minimum spanning trees and mutations for mice immunized with mRNA-LNP or adjuvanted protein.

Minimum spanning trees of D1D2 sequences, similar to those presented in Fig. 5c, for additional mice immunized by mRNA-LNP every two or four weeks (M3 through M10), or by adjuvanted protein every two weeks (M16 through M24). M6 and M7 were combined into one sample. Each tree presents the inferred lineage and all amino-acid mutations found in each mouse. The central black dot represents the inferred ancestral sequence which corresponds to the input sequence. Each circle indicates a distinct amino-acid sequence. Circle size is proportional to the number distinct nucleotide sequences with the same translation. Colored circles mark the translations encoded by the largest number of distinct sequences, with the rank order indicated by number. Branch length corresponds to evolutionary distance, defined as the number of amino-acid differences.

Extended Data Fig. 6.. Minimum spanning trees and mutations for mice immunized with mRNA-LNP or adjuvanted protein.

Minimum spanning trees of D1D2 sequences, similar to those presented in Fig. 5c, for additional mice immunized by mRNA-LNP every two or four weeks (M3 through M10), or by adjuvanted protein every two weeks (M16 through M24). M6 and M7 were combined into one sample. Each tree presents the inferred lineage and all amino-acid mutations found in each mouse. The central black dot represents the inferred ancestral sequence which corresponds to the input sequence. Each circle indicates a distinct amino-acid sequence. Circle size is proportional to the number distinct nucleotide sequences with the same translation. Colored circles mark the translations encoded by the largest number of distinct sequences, with the rank order indicated by number. Branch length corresponds to evolutionary distance, defined as the number of amino-acid differences.

Extended Data Fig. 7.. Neutralization of CD4-Ig variants with single or double mutations against a panel of HIV-1 isolates.

a Neutralization curves of CD4-Ig-v0 variants modified with R59K (orange), K90R (red), or combined (blue) against several HIV-1 PV. All curves were fitted with a variable slope four parameters dose response model. IC₅₀ values are presented in Fig. 6a. b Neutralization potency of WT CD4-Ig, unmodified or altered to include the most potent substitutions (A55V, N66S) identified from two earlier phage-display studies, or with R59K and R59K/K90R identified here, were compared using the indicated isolates. Isolates that were not neutralized were assigned a value of 100 μg/ml. Statistical significance was determined by two-way ANOVA with H-Šídák’s multiple comparisons (WT vs. WT-A55V, ^nsp = 0.4027; WT vs. WT-R59K, *p = 0.0126; WT vs. WT-N66S, ^nsp = 0.6446; WT vs. WT-R59K K90R, *p = 0.012; WT-A55V vs. WT-N66S, *p = 0.4027; WT-R59K vs. WT-N66S, ***p = 0.0003; WT-N66S vs. WT-R59K K90R, ***p = 0.0003). c Neutralization curves of CD4-Ig N30H, a recurring mutation from NGS analysis, against three HIV-1 PVs. d Neutralization potency of CD4-Ig variants against BG505 and TRO11 in an initial screening for potent CD4-Ig variants.

Extended Data Fig. 8.. Naturally occurring D1D2 variants improved the neutralization potency of CD4-Ig-v0.

a Representative neutralization curves of CD4-Ig-v0 and three naturally emerging variants against a 12-isolate global panel of HIV-1 pseudoviruses. These variants represented the nodes with the largest number of progenies identified in M1, M3, and M6/7 in Extended Data Fig. 6. Curves were fitted with a variable slope four parameters dose response model.

Extended Data Fig. 9.. Affinity matured CD4-Ig variants retained bioavailability.

a The polyreactivity of CD4-Ig variants were measured by immunofluorescence assays using HEp-2 cells and 200 μg/ml of each antibody. The autoreactive antibody 2F5 served as a positive control (pc). Baseline (dashed line) was determined as the fluorescence intensity of negative human serum. Each bar is an average of four independent measurements. Error bars indicate SEM. All CD4-Ig variants were significantly less polyreactive than 2F5, as determined by two-way ANOVA with Dunnett’s multiple comparison (*p = 0.0282, ****p < 0.0001). b Thermostability of CD4-Ig variants. Measured by differential scanning fluorimetry, each bar represents an average of two independent experiments. Significance was determined by two-way ANOVA with Dunnett’s multiple comparison (****p < 0.0001). c A summary of half-lives from two pharmacokinetic (PK) studies of CD4-Ig variants in immunocompromised hFcRn mice. Each dot represents a calculated half-life from one hFcRn transgenic mouse (n = 12 each group). Half-life was calculated by fitting a one-phase model. Significance was determined two-way ANOVA with Šídák’s multiple comparisons (*p = 0.0185, ****p < 0.0001). d Change of CD4-Ig serum concentration from a representative PK study. 8 mg/kg of the indicated CD4-Ig variants was infused intravenously into six nine-week-old mice per group. Sera were collected on days 1, 3, 6, 14, 21 and 31. CD4-Ig concentration was measured by ELISA with anti-CD4 antibodies.

Fig. 1:. Engineering primary murine B cells to express a B-cell receptor with CD4 domains 1 and 2.

a A representation of an engineered BCR with a potency and half-life enhanced form of CD4 domains 1 and 2 (D1D2) fused through a (G₄S)₃ linker to the amino-terminus of the heavy-chain variable region of the mouse antibody OKT3 (D1D2-OKT3-V_H). The OKT3 heavy chain pairs with an endogenous mouse light chain. b Introducing D1D2-OKT3-V_H at the murine heavy-chain locus. The CRISPR effector protein Mb2Cas12a targets the J4 coding region 5’ of a CTTA PAM, as represented. An rAAV-delivered homology directed repair template (HDRT) complements the 5’ UTR of a VH segment and the intron 3’ of JH4 using 576 bp and 600 bp homology arms, respectively. The edited genome replaces the VDJ-recombined heavy chain with a cassette encoding D1D2-OKT3-V_H. c Expression of D1D2-OKT3-V_H in primary mouse B cells. Expression of D1D2 in edited cells was measured by flow cytometry with monomeric HIV-1 gp120. Representative flow cytometry plots of B cells edited with HDRT targeting the 5’ UTR of VH1–34 or V1–64 were generated 48 h after electroporation. HDRT were delivered with rAAV transduced at 10⁴ multiplicity of infection (MOI). Controls include cells electroporated with Mb2Cas12a ribonucleoproteins (RNP) without rAAV (No HDRT) or without gRNA but transduced with HDRT-encoding rAAV (No gRNA). Plots were gated on viable singlet B cells. d Quantitation of editing efficiency in c from independent experiments. Each dot represents an average of two technical replicates in an experiment. Error bars represent standard error of mean (SEM). Statistical significance was determined by two-way ANOVA followed by H-Šídák’s multiple comparisons (****p < 0.0001, ^nsp = 0.8794). a, b are created with BioRender.com.

Fig. 2:. Engineered B cells generated neutralizing responses in immunized mice.

a Schedule of immunization and blood collections from mice analyzed in subsequent figures. Naïve B cells from CD45.1 donor mice were engineered ex vivo and 5 million cells per mouse were adoptively transferred to CD45.2 recipient mice 24 h later. Mice were immunized with SOSIP-TM (16055-ConM-v8.1) mRNA-LNP on two-week (2 wk, M1-M5, Day 2, 16, and 30, n = 5) or four-week (4 wk, M6-M10, Day 2, 30, 58, n = 5) intervals, and serum was collected seven days after each immunization. Spleens and lymph nodes were harvested four days after the final immunization and B cells isolated from these tissues were analyzed by flow cytometry and next-generation sequencing (NGS). b, c Neutralizing responses of sera from mice immunized with SOSIP-TM mRNA. Sera from mice immunized at two-week (b) or four-week (c) intervals were measured individually after prime (1^st) and boosts (2^nd and 3^rd) for their ability to neutralize BG505 HIV-1 pseudovirus in TZM-bl cell assays. Sera from mice (n = 3) engrafted with unedited CD45.1 B cells and immunized on two-week interval (grey dots) served as mock controls. 100 μg/ml CD4-Ig-v0 combined with normal mouse serum served as a positive control. Dots and error bars indicate median and interquartile range for each group. d A summary of the 50% inhibitory dilutions (ID₅₀) of sera from each immunized mouse in b and c. Statistical significance was determined using repeated-measures two-way ANOVA with Geisser-Greenhouse correction (2 wk interval 1^st vs. 2^nd and 3^rd, *p = 0.0119; 4 wk interval 1^st vs. 2^nd and 3^rd, *p = 0.029). e Serum concentration of CD4-OKT3-IgG after each immunization, measured by ELISA with an anti-CD4 antibody and CD4-OKT3-IgG as the standard.

Fig. 3:. Engineered B cells persisted in vivo following immunization.

a Quantification of CD45.1-positive donor B cells in vaccinated mice. Four days after the final vaccination of mice (M1-M10) characterized in Fig. 2, B cells were isolated from their lymph nodes and spleens. B cells were analyzed by flow cytometry. Figure shows the percent of CD45.1-positive donor cells in mice immunized at two-week (2 wk, M1-M5) or four-week intervals (4 wk, M6-M10). Mock mice (n = 3) similarly engrafted with unedited cells and immunized in parallel every two weeks served as the controls. For gating strategies, see Supplementary Fig. 2a. (2 wk vs. mock, ^nsp = 0.3943; 4 wk vs. mock, ^nsp = 0.7762) b A greater proportion of CD45.1 donor cells binding to HIV-1 gp120. The cells analyzed in panel a were measured for binding to HIV-1 gp120. See Supplementary Fig. 2b for source flow cytometry analysis. (**p = 0.0033, ***p = 0.0007, ^nsp = 0.9512) c Enrichment of gp120-binding donor cells in the germinal center. Germinal center (GC) B cells (CD19+ CD138_low CD38− GL7+) were analyzed by flow cytometry for their ability to bind gp120. Two representative mice (M1 and M2) immunized at two-week interval are shown. Additional examples (M7 and M9) are provided in Extended Data Fig. 3d. d Quantification of results from GC staining experiments (M1, M2, M7 and M9, n = 4, **p = 0.015). e Distribution of isotypes among D1D2-expressing donor B cells sorted in panel a-b and compared with edited donor B cells before engraftment, as determined by NGS. Primers specific to the human CD4 was used to construct cDNA library from CD45.1-positive cells binding to gp120. Fewer B cells were isolated from M6 and M7, and these cells were therefore combined for efficient sequencing. For a, b, d, error bars indicate standard deviation. Statistical significance was determined by generalized linear mixed model followed by Tukey HSD pairwise comparisons.

Fig. 4:. D1D2-expressing B cells hypermutated and class switched in vivo.

a Nucleotide mutation frequency across the D1D2-encoding region. mRNA isolated from CD45+ IgG+ gp120-binding B cells from each of the 10 mice was analyzed by NGS and the mean frequency of nucleotide mutations is plotted for each position. The dash line represents the average mutation frequency per nucleotide across M1-M10. Triangles represent the most frequent coding mutations. Codons for R59 and K90 are indicated. b Frequency distribution of synonymous (Syn) and non-synonymous (Non-syn) mutations within D1D2. Nucleotide changes were classified based on whether they were synonymous and non-synonymous. Each dot represents the mutation frequency at one nucleotide position, averaged on five mice in each group. The top two dots indicate the nucleotide mutations leading to R59 and K90 mutations. A cut off of 0.001% for the mutation frequency was applied. Line indicates the median. c Average number of accumulated nucleotide mutations per unique sequence. Significance was determined by two-tailed unpaired t test (^nsp = 0.6384). Error bars represent SEM. d The frequency of synonymous mutations within domains 1 (D1) and 2 (D2) for mice immunized at two-week and four-week intervals. The center line indicates mean, and boxes denote quartile range. Repeated measure mixed effects analysis with H-Šídák’s multiple comparisons (**p = 0.0095, ^nsp = 0.101). e Distribution of accumulated nucleotide mutations per unique D1D2 sequence. The center line indicates mean, and boxes denote quartile range of data. Statistical significance in b, e was determined by mixed effects analysis with H-Šídák’s multiple comparisons (**p = 0.0079, ****p < 0.0001).

Fig. 5:. Diverse and convergent amino-acid mutations in engrafted mice.

a The frequency of amino acid changes across D1D2 sequences from each mouse (M1 through M10). Three amino acids with the highest mutation rate are labeled. M6 and M7 were combined in sequencing analysis. b Amino acid changes found in the eight most frequently mutated D1D2 residues from the indicated mice immunized at two-week intervals (M1, M2) are represented. The data for remaining mice are shown in Extended Data Fig. 5. c Minimum spanning trees of D1D2 sequences from individual mice. Each tree presents the inferred lineage and all amino-acid mutations found in M1 and M2. The central black dot represents the inferred ancestral sequence which corresponds to the input sequence. Each circle indicates a distinct amino-acid sequence. Circle size is proportional to the number distinct nucleotide sequences with the same translation. Coloured circles mark translations encoded by the eighteen largest number of distinct sequences, with the rank order indicated by number. Branch length corresponds to evolutionary distance, defined as the number of amino-acid differences. The figures for the remaining mice are provided in Supplementary Fig. 5.

Fig. 6.. In vivo hypermutations in D1D2 improve the neutralization potency of CD4-Ig-v0.

a Neutralization potency of CD4-Ig-v0 and its variants modified with R59K, K90R, or combined against the indicated isolates in TZM-bl assays. Curves are shown in Supplementary Figure 6a. Statistical significance was determined by two-way ANOVA with Dunnett’s multiple comparisons (***p = 0.0004, ****p < 0.0001). b Location of selected D1D2 mutations (lime) shown on a structure of CD4 (gold) bound to an HIV-1 Env trimer (light grey). Structure was adapted from pdb:5U1F. c Representative neutralization curves of CD4-Ig. CD4-Ig-v0 and v1-v4 variants bearing combinations of mutations listed below the figure, against a 12-isolate global panel of HIV-1 pseudoviruses in TZM-bl assays. All curves were fitted with a variable slope four parameters dose response model. The table details the mutations present in each CD4-Ig variant. d IC₅₀ of CD4-Ig. CD4-Ig-v0 and the indicated engineered (v1-v4) and naturally emerging variants (M1–2, M3–1, M6/7–1) against the global panel (****p < 0.0001). Each dot represents an average of two independent experiments. Center lines indicate median. Isolates that were not neutralized were assigned a value of 100 μg/ml. Statistical significance in d was determined by repeated-measures two-way ANOVA with Dunnett’s multiple comparisons.

Fig. 7.. CD4-Ig variants bind Env trimers with higher affinity than CD4-Ig-v0.

a Representative SPR sensorgrams of binding kinetics between CD4-Ig variants and 16055-ConM-v8.1 SOSIP trimers. Anti-human IgG antibodies were immobilized on the surface to capture CD4-Ig. Single-cycle kinetics was performed after CD4-Ig capture by injecting SOSIP trimers at five concentrations (800, 400, 200, 100, 50 nM) followed by one long dissociation step at 25°C and reported with response units (RUs). Experimental data are shown in red. Fitted curves calculated using 1:1 kinetics Langmuir binding model are shown in black. Kinetic rate constants are displayed as an average ± standard deviation of two or three independent experiments.