Discovery of VH domains that allosterically inhibit ENPP1

Abstract

Ectodomain phosphatase/phosphodiesterase-1 (ENPP1) is overexpressed on cancer cells and functions as an innate immune checkpoint by hydrolyzing extracellular cyclic guanosine monophosphate adenosine monophosphate (cGAMP). Biologic inhibitors have not yet been reported and could have substantial therapeutic advantages over current small molecules because they can be recombinantly engineered into multifunctional formats and immunotherapies. Here we used phage and yeast display coupled with in cellulo evolution to generate variable heavy (VH) single-domain antibodies against ENPP1 and discovered a VH domain that allosterically inhibited the hydrolysis of cGAMP and adenosine triphosphate (ATP). We solved a 3.2 Å-resolution cryo-electron microscopy structure for the VH inhibitor complexed with ENPP1 that confirmed its new allosteric binding pose. Finally, we engineered the VH domain into multispecific formats and immunotherapies, including a bispecific fusion with an anti-PD-L1 checkpoint inhibitor that showed potent cellular activity.

Extended Data Figure 1.. Sequences, cellular binding, and stability of VH panel.

A,B. Schema of phage-display selection and elution by TEV protease treatment (A) or 10 mM ATP treatment (B). C. CDR H1-3 sequences for VH panel. D. Representative flow cytometry histograms used to generate bar graph summarizing fold-changes for binding to OS384 SG and ENPP1 KO in Figure 1D. E. SEC traces for VH24-Fc, VH27-Fc, VH31-Fc, and VH38-Fc.

Extended Data Figure 2.. AHEAD yeast display campaign to affinity mature VH27.

A. FACS gating used for sorting less than 1% AHEAD yeast cell population based on expression (HA tag) and antigen binding. In round 4 the fluorophores were switched (SA-488 and anti-HA-647). B-D. Biolayer interferometry signals and fits to determine K_D, k_association, and k_dissociation for VH27-Fc with scaffold mutations T75I (B), A89V (C), and T75I/A89V (D). Data are representative of two independent experiments. E,F. Mouse ENPP1 (mENPP1) was treated with 500 nM VH27/T75I/A89V-Fc, Fc isotype, or PBS to evaluate cross-reactivity using ATP (E) and cGAMP (F) substrates. Bar graphs reports mean and SEM for N=3 independent replicates and statistics were calculated using two-tailed Student’s t-test. G. Dose titration of VH27/T75I/A89V-Fc or indicated controls were tested for ENPP1 inhibition in C57BL/6J mouse plasma supplemented with 1 mM cGAMP for 90 min (N=3 mice, 1 experiment per mouse).

Extended Data Figure 3.. Additional data for multivalent constructs.

A,B. SEC traces for biparatopic inhibitor (A) and bispecific inhibitor (B). C. Representative flow cytometry histograms used to generate bar graph summarizing fold-changes for binding to OS384 SG and ENPP1 KO in Figure 3F. D. Representative flow cytometry histograms for NFAT-GFP Jurkat activation assays when treated with indicated concentrations of ENPP1-coupled beads or no beads and 10 nM BiTE or PBS. E. Unformatted immunoblot with visible ladder used for image in Figure 5F.

Extended Data Figure 4.. Additional data for cryo-EM data collection and processing.

A. Representative unprocessed micrograph (N=4426). B. Single-particle cryo-EM image processing workflow. C. ENPP1-VH complex colored by local resolution. D. Orientation angle distribution of particles in the final reconstruction. E. Gold-standard FSC of the final non-uniform refinement. (resolution = 3.2Å at gold-standard FSC=0.143). F. FSC curves between the experimental map and model. G. Histogram and Directional FSC Plot from 3DFSC. H. Experimental density for ligand in ENPP1 active site (mesh) with AMP modeled in (green sticks) and two zinc atoms (gray spheres).

Extended Data Figure 5.. Additional data for structural and mechanistic analyses.

A-D. Validation of VH-ENPP1 complex. Complex was purified by SEC (A) and co-elution of VH and ENPP1-Fc was analyzed by SDS-PAGE gel (representative of N=3 independent experiments) (B). Mass photometry of antigen alone (C) and SEC-eluted complex peak (D) demonstrated mass shift of approximately 30 KDa suggesting 2:1 VH:antigen stoichiometry. E. Cryo-EM structure of VH-ENPP1 complex aligned to PDB 6wjf. F. Cryo-EM structure of VH-ENPP1 complex aligned to PDB 4gtw with sequence alignment of human and mouse ENPP1. Residues colored red were within 5Å of VH CDR H1, H2, or H3. G. Close-up of clashing with pApG when VH-ENPP1 complex (cyan) is aligned to PDB 6aek (yellow). H. Representative biolayer interferometry signals and fits for VH27.2-Fc Y102A (N=2 independent replicates). I,J. Michaelis-Menten kinetic analysis for VH27.2-Fc Y102A for ATP (I) and cGAMP (J) (mean for N=2 independent replicates). K. Representative biolayer interferometry signals and fits for VH27.2-Fc W104A (N=2 independent replicates). L,M. Michaelis-Menten kinetic analysis for VH27.2-Fc W104A for ATP (L) and cGAMP (M) (mean for N=2 independent replicates).

Fig. 1.. Phage display generated high affinity VH domains recognizing native ENPP1 on PDX-derived osteosarcoma cells, and VH27-Fc inhibited ATP and cGAMP hydrolysis.

A. Structure of extracellular domain of ENPP1 and recombinant ENPP1 C-terminal Fc-fusion antigen for phage display (ENPP1-Fc). B. Representative biolayer interferometry signals and fits for each VH-Fc binding ENPP1-Fc antigen or Fc-biotin control. C. Table of K_D values (mean for N=2). D. VH-Fc binding to PDX-derived OS384 cell line engineered with ENPP1 knockout (KO) or safe-guide (SG). The bar graph reports mean and SEM of the fold-change (SG/KO) in median fluorescent signal (N= 3 or 4 independent replicates). Statistics were calculated using one-tailed Student’s t-test. E, F. Michaelis-Menten kinetics were determined for VH27-Fc using ATP (E) or cGAMP (F) substrates (mean and standard deviation for N=3 independent replicates). G. VH27-Fc inhibited secreted ENPP1 in ex vivo plasma supplemented with 1mM cGAMP over 90 min and 24 hr time courses. Fc isotype treatment and condition with no cGAMP added (No cGAMP) were included as controls (N=3 for Donor 1).

Fig. 2.. Affinity maturation of VH27 improved affinity, inhibitory potency, and stability.

A. Schema of AHEAD yeast display selection rounds. B. T75I and A89V NGS read frequencies and ranks. C. Association and dissociation rates and affinity constants for VH27-Fc, VH27/T75I-Fc, VH27/A89V-Fc, and VH27/T75I/A89V-Fc binding to ENPP1-Fc antigen were determined by biolayer interferometry (mean and standard deviation for N=2 independent replicates). D,E. Michaelis-Menten enzyme kinetics for VH27/T75I/A89V-Fc using ATP (D) and cGAMP (E) substrates (mean and standard deviation for N=3 independent replicates). F. Human plasma supplemented with 1 mM cGAMP was treated with a titration of VH27/T75I/A89V-Fc for 90 min and cGAMP degradation was measured. Fc isotype treatment and condition with no cGAMP added (No cGAMP) were included as controls (N=3 for each donor). G. Differential scanning fluorimetry was used to measure T_m of VH27 (WT) and VH27/T75I/A89V as single-domain VHs in non-reducing and reducing (6.25% BME) conditions. Bar graph reports mean and SEM for N=4 or 5 independent replicates and statistics were calculated using two-tailed Student’s t-test. H. VH27-Fc and VH27/T75I/A89V-Fc were analyzed by SEC and peak area ratios between ‘aggregate peak’ and ‘Fc peak’ were analyzed. Bar graph reports mean for N=2 or 3 independent replicates.

Fig. 3.. VH27.2 was formatted into bi-paratopic and bispecific constructs.

A. Structures of bi-paratopic and bispecific tetravalent Fc inhibitors. B,C. Biolayer interferometry mapping the epitopes of the VH panel with respect to the VH27 epitope when VH27-Fc was pre-loaded on the sensor (B), and when VH24-Fc, VH31-Fc, or VH38-Fc was pre-loaded on the sensor (C). D. Representative biolayer interferometry signals and fits for bi-paratopic VH27.2/VH31 inhibitor binding ENPP1-Fc antigen or Fc-biotin control (mean and standard deviation for N=2 independent replicates). E. Bi-paratopic molecule inhibited ENPP1 in human plasma (1mM cGAMP, 90 min). Fc isotype treatment and condition with no cGAMP added (No cGAMP) were included as controls (N=2 independent replicates for each donor). F. Binding of VH27.2-Fc and bi-paratopic molecule to OS384 SG and ENPP1 KO cells. The bar graph reports mean and SEM of the fold-change (SG/KO) in median fluorescent signal (N=5 or 6 independent replicates). Statistics were calculated using one-tailed Student’s t-test. G,H. Representative biolayer interferometry signals and fits for bispecific inhibitor binding ENPP1-Fc antigen (G), PDL1-Fc antigen (H), and Fc-biotin control (mean and standard deviation for N=2 independent replicates). I. Bispecific molecule inhibited ENPP1 in human plasma (1mM cGAMP, 90 min). Fc isotype treatment and condition with no cGAMP added (No cGAMP) were included as controls (N=2 independent replicates for each donor).

Fig. 4.. Bi-paratopic and bispecific constructs improved localization on tumor cells and increased inhibition of ENPP1 on cell membranes.

A. MDA-MB-231 cells were stained with a titration of VH27.2-Fc, VH31-Fc, bi-paratopic molecule, bispecific molecule, or Fc-Envafolimab and median fluorescence intensity values were used to fit the EC₅₀ curve (mean for N=2 or 3 independent replicates). B. MDA-MB-231 cells were treated with cGAMP and the indicated concentration of VH27.2-Fc, bi-paratopic inhibitor, or bispecific inhibitor. cGAMP remaining in the media was measured by cGAMP ELISA. Data were normalized to the range between the media treated without cells and cells treated with PBS, as shown in C. Fc isotype treated at 3 uM was also included as a control in C. D. MDA-MB-231 cells were treated with pNP-TMP and the indicated concentration of VH27.2-Fc, bi-paratopic inhibitor, or bispecific inhibitor and pNP-TMP hydrolysis was measured. Data were normalized to the PBS condition presented in E. Fc isotype treated at 3 uM was also included as a control in E. Data in B-E represent mean and SEM for N = 3-5 biological replicates.

Fig. 5.. VH27.2 was recombinantly engineering into immunotherapy scaffolds and next-generation protein degraders.

A. Representative biolayer interferometry signals for bivalent VH27.2-Fc, tetravalent bispecific Fc, and single domain VH27.2 (no Fc) binding to CD16-Fc antigen or Fc-biotin control (N=2). B. Structure of bispecific T cell engager (BiTE) combining VH27.2 with arm recognizing CD3 (OKT3 scFv). C. Jurkat cells expressing NFAT-GFP reporter were incubated with beads coated with Fc-ENPP1 or a no bead control, and were treated with 10 nM BiTE or PBS control. GFP expression driven by NFAT activation was measured by flow cytometry. Data were normalized to the no bead/no BiTE condition. Bar graph reports mean and SEM for N=3 independent replicates. Statistics were calculated using two-tailed Student’s t-test. D. Structure of ‘knob-into-hole’ bispecific AbTAC degrader combining anti-ENPP1 VH with RNF43-recruiting IgG arm. E, F. MDA-MB-231 cells were treated with PBS or a dose titration of VH27.2 AbTAC, and ENPP1 levels were measured by immunoblot. Cells were also treated with 500 nM VH27.2-Fc and Fc isotype controls. ENPP1 densities were normalized to ACTIN loading control. The percent of ENPP1 remaining relative to the PBS treatment was calculated. E. Graph summarizes the mean and SEM for N=3-5 independent replicates. Statistics were calculated using two-tailed Student’s t-test. F. A representative immunoblot for one experiment. G. MDA-MB-231 cells were pre-treated with VH27.2 AbTAC, indicated control molecules, or PBS for 24 hr. After removing the molecules and washing the cells, pNP-TMP hydrolysis was measured. Data were normalized to PBS condition and graph represents mean and SEM for N= 3 independent replicates. Statistics were calculated using two-tailed Student’s t-test.

Fig. 6.. Cryo-EM reveals VH binding ENPP1 proximal to the catalytic site.

A. Cryo-EM 3D reconstruction of VH27.2 bound to ENPP1 ectodomain. B. View of the CDR H1 epitope. C. View of the CDR H2 epitope. D. View of the CDR H3 epitope. E. Table summarizing interactions between VH and ENPP1 residues. F. Biolayer interferometry comparing binding kinetics of VH27.2-Fc alanine variants. Traces are representative of two independent experiments. G. Inhibitory potencies of VH27.2-Fc alanine variants relative to WT VH27.2-Fc treated at 500 nM for ATP, pNP-TMP, and cGAMP substrates. Bar graph reports mean for two independent replicates. H. Biolayer interferometry comparing affinity of VH27.2-Fc to WT, K528A, F346A, and H380A ENPP1-Fc. Traces are representative of two independent experiments. I. Biolayer interferometry comparing binding kinetics of VH27.2-Fc phenylalanine variants. Traces are representative of two independent experiments. J. Inhibitory potencies of VH27.2-Fc phenylalanine variants relative to WT VH27.2-Fc treated at 500 nM for ATP, pNP-TMP, and cGAMP substrates. Bar graph reports mean for two independent replicates. K. ATP/cGAMP K_i fold-change value for WT, Y102A, and W104F VH27.2 -Fc. L. Linear regression and R² values correlating K_D and K_i values for ATP and cGAMP substrates for WT, Y102A, and W104F VH27.2-Fc variants.