Novel Immunomodulatory Proteins Generated via Directed Evolution of Variant IgSF Domains

Abstract

Immunoglobulin superfamily member (IgSF) proteins play a significant role in regulating immune responses with surface expression on all immune cell subsets, making the IgSF an attractive family of proteins for therapeutic targeting in human diseases. We have developed a directed evolution platform capable of engineering IgSF domains to increase affinities for cognate ligands and/or introduce binding to non-cognate ligands. Using this scientific platform, ICOSL domains have been derived with enhanced binding to ICOS and with additional high-affinity binding to the non-cognate receptor, CD28. Fc-fusion proteins containing these engineered ICOSL domains significantly attenuate T cell activation in vitro and in vivo and can inhibit development of inflammatory diseases in mouse models. We also present evidence that engineered ICOSL domains can be formatted to selectively provide costimulatory signals to augment T cell responses. Our scientific platform thus provides a system for developing therapeutic protein candidates with selective biological impact for treatments of a wide array of human disorders including cancer and autoimmune/inflammatory diseases.

Keywords: ICOS ligand (ICOSL); IgSF; anti-inflammatory; costimulation; protein engineering; protein therapeutics.

Figure 1

Our scientific platform can yield yeast population outputs with increased affinity for cognate ligands and/or new binding partners. (A) Selections with two ligands results in yeast display of engineered ICOSL domains with increased binding toward both counter-structures. Yeast were transformed with a 1st generation random engineered ICOSL domain library and affinity matured by selection with indicated recombinant receptors. Individual receptor binding to bulk yeast populations by flow cytometry are shown. Improvements to binding against both receptors were noted at each selection. (B) Schematic showing yeast display strategy for selection and characterization of engineered vIgDs with altered specificity and/or affinity.

Figure 2

Directed evolution of ICOSL by yeast display leads to recombinant vIgD hits with enhanced binding to ICOS and CD28. ICOSL DNA sequences derived from yeast display outputs were cloned into a mammalian expression vector and produced in HEK-293 cells. Engineered ICOSL vIgD-Fc were subsequently titrated on CD28 (left) and ICOS (right) transfectants. Binding was detected by flow cytometry and Mean Fluorescence Intensity (MFI) was plotted vs. protein concentration. For comparison, ICOSL vIgD-Fc derived from 1st generation (A160), 2nd generation (A2237), and 3rd generation (A3256) are compared to WT ICOSL, showing progressive increases in binding for each target.

Figure 3

ICOSL mutations identified at predicted interaction sites of homologous B7/CD28 family members. (A) Interaction of human CD80 with human CTLA4 as deduced from the crystal structure of CD80 extracellular domain dimer complexed with single extracellular domain of CTLA4 per CD80 monomer (PDB ID: 1I8L). (B) Homology model of human ICOSL extracellular domain based on CD80/CTLA4 complex. ICOSL residues predicted to be located within 5 angstroms of ICOS are colored cyan and marked by an asterisk. ICOSL residues frequently mutated through yeast library selection are colored purple and indicated in bold. Residues are split between front and back view of ICOSL dimer (rotated 180°) to maximize visibility in figure.

Figure 4

Soluble ICOSL vIgD-Fc effectively attenuate T cell responses in mixed lymphocyte reaction (MLR). Monocyte derived dendritic cells from one donor were used to allogeneically stimulate purified T cells from another donor. Responses were measured by assessing (A) CD4 T cell proliferation, (B) CD8 T cell proliferation, and (C) IFNγ levels in culture supernatants. Proliferation results use CFSE dilution to show the percentage of divided cells vs. protein concentration. IFNγ levels are shown as pg ml⁻¹ vs. protein concentration, with IC₅₀ pM values calculated using GraphPad Prism and shown in parenthesis on graph label. ND, not determined. Results shown are representative of at least three separate experiments performed with each protein.

Figure 5

ICOSL vIgD-Fc suppress immune responses in vivo. (A) A delayed type hypersensitivity (DTH) model was performed by sensitizing mice with OVA and subsequently rechallenging with OVA in the ear pinna. Groups of seven mice treated with either abatacept or ICOSL vIgD-Fc showed significantly less OVA-induced ear swelling as compared to PBS treated animals (^**p < 0.0001 by 1-way ANOVA). Bars shown are the group mean (s.d.). (B,C) An acute model of graft-versus-host-disease (GvHD) was performed by adoptively transferring human PBMC into immunodeficient NSG mice (n = 9/group). (B) High affinity ICOSL vIgD-Fc significantly prolonged survival and (C) significantly reduced mean disease activity index (DAI). Administration of ICOSL vIgD-Fc protected from effects of GvHD at levels comparable to or better than belatacept, but wild-type ICOSL-Fc or lower affinity ICOSL vIgD-Fc were not effective in protecting from GvHD in this model. For (B), by log-rank test, belatacept and high affinity ICOSL vIgD-Fc significantly prolonged survival as compared to saline and WT ICOSL-Fc (p < 0.001) treatments; ICOSL vIgD-Fc A2237 prolongs survival as compared to belatacept (p = 0.065). For (C), by 2-way repeated-measures ANOVA, belatacept and high affinity ICOSL vIgD-Fc significantly reduce DAI scores as compared to saline and wild-type ICOSL-Fc (p < 0.001); ICOSL vIgD-Fc A2229 and A2237 were significantly better at reducing DAI scores than belatacept (p = 0.053 and p = 0.035 for A2229 and A2237, respectively). Both studies were performed at least twice with representative experiments shown here.

Figure 6

Immobilized, engineered ICOSL-Fc variants costimulate T cells when plated with sub-optimal anti-CD3. Human T cells were stimulated with a range of anti-CD3 concentrations and a constant concentration of coimmobilized ICOSL-Fc variants (40 nM). (A) CD4 and (B) CD8 T cell proliferation was monitored by CFSE dilution. Proliferation is reported as percentage of T cells divided vs. anti-CD3 concentration. (C) IFNγ production was measured by ELISA in supernatants collected at 72 h and plotted as pg/ml cytokine vs. anti-CD3 concentration. Each point represents the mean of triplicate determinations ±s.d.

Figure 7

ICOSL vIgD-Fc expressed on the surface of cells deliver a costimulatory T cell signal. Cell Trace Far Red labeled human T cells were cocultured with K562 cells expressing the indicated versions of ICOSL vIgD-Fc and stimulated with a range of soluble CD3 antibody concentrations. Cells were harvested after 72 h and proliferation of (A) CD4 and (B) CD8 human T cells is reported as percent of cells divided vs. anti-CD3 concentration. Each point represents the mean of triplicate wells with error bars showing standard deviation (s.d.).

Figure 8

ICOSL vIgD proteins can be formatted to provide localized T cell costimulation. (A) Schematic diagram of a tumor localizing fusion protein consisting of an ICOSL vIgD costimulatory domain (N-terminal), a variant NKp30 domain to localize the protein to B7H6 expressing tumors, and an antibody Fc domain. (B–D) ICOSL-NKp30 vIgD-Fc fusion proteins can provide T cell costimulatory signals that are dependent on the presence of B7H6. Plates were coated with 40 nM recombinant B7H6-Fc protein and 10 nM anti-CD3. CFSE-labeled primary human T cells were added with titrated concentrations of the variant ICOSL-Fc alone (orange circle), the variant NKp30-Fc alone (purple circle), or an ICOSL-NKp30-Fc variant fusion protein (green circles) for 3 days. Cells were analyzed for proliferation by CFSE dilution in (B) human CD4 or (C) human CD8 T cells. (D) Supernatants were collected and assessed for IFNγ production by ELISA. (E) Variant ICOSL-NKp30-Fc fusion proteins can also be used to localize T cell costimulatory signals to B7H6 positive cells. K562 cells that express B7H6 were plated with CFSE-labeled human T cells and varying concentrations of a control Fc-protein (red circles), the variant NKp30-Fc protein alone (purple circles), the variant ICOSL-Fc fusion protein alone (orange circles), a wild type NKp30-ICOSL-Fc fusion protein (blue circles) or an ICOSL-NKp30-Fc fusion variant (green circles). Results shown are representative of at least two experiments, and individual points represent mean values of triplicate wells ±s.d.

Figure 9

Engineered ICOSL costimulatory vIgDs can be fused to an antibody, retain binding, and provide localized costimulatory signals. (A) Schematic diagram of V-mAb fusion proteins. Blue structures represent heavy and light chains of the monoclonal antibody, solid red structures represent the IgV domain of an ICOSL vIgD, and open red structures are the IgC domain of an ICOSL vIgD. ICOSL-trastuzumab V-mAbs retain binding to (B) HER2, (C) CD28, and (D) ICOS. HEK-293 HER2 transfectants were stained with titrated amounts of indicated V-mAbs and analyzed by flow cytometry. (E) ICOSL-trastuzumab V-mAbs retain costimulatory activity. Titrated amounts of indicated V-mAbs were coated to plates with a fixed concentration of anti-CD3 antibody (10 nM) and incubated for 3 days with CFSE labeled T cells. Graph shows the percentage of CD8+ T cells that divided vs. protein concentrations. (F) ICOSL-trastuzumab V-mAbs costimulate primary human T cells in the presence of a HER2+ tumor cell line. NCI-N87 (HER2+) human gastric carcinoma cells were transduced with anti-CD3 single chain Fv (OKT3). Primary human T cells were plated with OKT3 expressing tumor cells at an E:T ratio of 1:4 and assayed 72 h later for IFN-γ production. (G,H) ICOSL-trastuzumab V-mAb driven T cell costimulation is dependent on HER2 expression. Primary human T cells were transduced with lentiviral vector encoding the HLA-A^*0201 restricted TCR directed against the viral oncoprotein human papillomavirus type 16 (HPV-16) E6. Parental T2 cells (G) or HER2-transduced T2 cells (H) were pulsed with 1 ng ml⁻¹ E6 peptide (29–38) for 90 min then plated with E6 TCR transduced T cells at an E:T ratio of 1:3. ICOSL-trastuzumab V-mAbs were added at the indicated concentrations and supernatants were harvested for assessment of IFNγ after 24 h. Localization of the ICOSL vIgD on HER2+ targets with the ICOSL-trastuzumab V-mAb significantly increased IFNγ production over trastuzumab alone.