Rational Design of a Bifunctional AND-Gate Ligand To Modulate Cell–Cell Interactions

Abstract

Protein “AND-gate” systems, in which a ligand acts only on cells with two different receptors, direct signaling activity to a particular cell type and avoid action on other cells. In a bifunctional AND-gate protein, the molecular geometry of the protein domains is crucial. Here we constructed a tissue-targeted erythropoietin (EPO) that stimulates red blood cell (RBC) production without triggering thrombosis. The EPO was directed to RBC precursors and mature RBCs by fusion to an anti-glycophorin A antibody V region. Many such constructs activated EPO receptors in vitro and stimulated RBC and not platelet production in mice but nonetheless enhanced thrombosis in mice and caused adhesion between RBCs and EPO-receptor-bearing cells. On the basis of a protein-structural model of the RBC surface, we rationally designed an anti-glycophorin–EPO fusion that does not induce cell adhesion in vitro or enhance thrombosis in vivo. Thus, mesoscale geometry can inform the design of synthetic-biological systems.

Graphical Abstract

Hormones and cytokines act on receptors on diverse cells to achieve coordinated biological responses. Some activities are desired, while others constitute drug side effects. For example, cortisol acts on the glucocorticoid receptor in immune cells and is anti-inflammatory but acts on other cells to increase blood sugar and induce bone resorption. Similarly, erythropoietin (EPO) stimulates red blood cell (RBC) production but also increases thrombotic activity. These responses are adaptive in the context of hemorrhage, but increased thrombotic activity may underlie the increased frequency of heart attacks and strokes seen in EPO-treated kidney failure patients.

We previously designed a class of AND-gate proteins, “chimeric activators”, in which a cytokine or hormone is mutated to reduce its activity and is attached to a targeting element that binds to another protein on a subset of receptor-bearing cells (Figure 1a–d). Chimeric activators are similar to other fusion proteins such as blinatumomab¹ or immunocytokines,² which connect one cell to another. Thus, implicit in this design is the potential to cross-link cells, whether desired or not. Understanding the structural determinants for engineered cell–cell interactions will be important in the design of next-generation CAR-T cells, bispecific T-cell engagers, and other fusion proteins.³

Figure 1.

(a) Protein sequence organization of a Targeted EPO chimeric activator. (b) The chimeric activator binds to a target cell via the targeting element and tethers the mutated activity element to the cell surface so that the mutated activity element binds to its receptor despite the mutation. (c) On a nontarget cell that lacks the surface marker to which the targeting element binds, the mutated activity element has little effect. (d) Targeted EPO molecules also bind to mature RBCs via GPA, creating a sink for a Targeted EPO to extend its plasma half-life. EPO-R is not present on these cells. (e) Extracellular portion of GPA (blue line), showing amino acid positions (numbers), O-linked (small arrows) and N-linked (large arrow) glycosylation sites, and the positions of the 10F7 and IH4 epitopes. (f) The 10F7-35-EPO(R150A) and IH4-5-EPO(R150A) forms of Targeted EPO, showing sizes in angstroms (Å) based on typical scFv and nanobody sizes, an extended conformation of the Gly/Ser linkers, and solved structures of EPO. Some images were created with BioRender.com.

We previously constructed an AND-gated EPO that activates EPO receptors (EPO-Rs) on RBC precursors but not on other cells.^4,5 “Targeted EPO” proteins consist of an antibody element that binds to glycophorin A (GPA), EPO with a mutation that reduces EPO activity, and a connecting linker (Figure 1a).^4,5 GPA is abundant on RBCs (800 000 copies/cell) and late RBC precursors (50 000 copies/cell).^4,6 GPA is not expressed on cells that control thrombosis and tumor angiogenesis or on solid tumor cells. EPO-Rs on late RBC precursors mediate EPO-dependent maturation but then disappear.^4,7 Targeted EPO was designed to bind to late RBC precursors and stimulate EPO-Rs on only those cells (Figure 1b). On nontarget cells it fails to activate EPO-Rs because the mutated EPO alone binds poorly to receptors (Figure 1c). Binding to GPA on mature RBCs is a desirable secondary consequence, as it extends the plasma half-life of the fusion protein (Figure 1d).⁴

The previously characterized form of Targeted EPO had a long linker between the antibody and EPO elements and an anti-GPA antibody element that binds to a membrane-distal epitope. That protein stimulated production of RBCs but not platelets, the latter being considered an indicator of off-target cell effects. While this result indicated cell-type specificity of action, platelet number alone is not strongly correlated with thrombotic propensity, and the possibility remained that the fusion protein might still cause thrombosis despite its seeming cell-type specificity. In this work, we found, surprisingly, that our initial fusion protein might have thrombotic activity in vivo. The structure–function analysis presented here indicates that, depending on the Targeted EPO configuration, the EPO moiety tethered to mature RBCs could interact with EPO-Rs on other cells and lead to cell adhesion but that the potential for such interaction could be eliminated by rational design. In vivo, a poorly designed fusion protein could bind to both RBCs and endothelial cells⁸ or immune cells such as macrophages⁹ and mediate undesired cell–cell adhesion or signaling.⁴

RESULTS AND DISCUSSION

The present work defines structural features of anti-GPA–EPO chimeric activators designed to act specifically in cis on GPA-expressing cells to avoid side effects that may arise from unwanted cell–cell interactions. Previously we characterized a form of Targeted EPO consisting of V regions from the 10F7 antibody, a 35-amino acid linker, and EPO with the weakening mutation R150A; this version is termed 10F7-35-EPO(R150A) (Figure 1f). This protein specifically stimulated production of RBCs but not platelets, the latter being considered an indicator of off-target cell effects. To directly assess the impact of this molecule on thrombosis, we performed a “bleeding time” assay in which the tail of a treated mouse was transected and the time to cessation of bleeding was measured.

Contrary to expectation, the 10F7-35-EPO(R150A) form of Targeted EPO promoted thrombosis in the bleeding time assay (Figure S1). We hypothesized two mechanisms by which this might occur. First, the antibody element could by itself promote blood clotting. Second, the RBC-bound Targeted EPO could cross-link RBCs with other cells bearing EPO-Rs, such as vascular endothelial cells or leukocytes, slowing blood flow. To determine whether either or both of these hypotheses were correct, we constructed a series of Targeted EPO variants testing anti-GPA antibody elements (10F7, 1C3, R18, and IH4), fusion protein configurations, and linker lengths.

Targeted EPO Activates EPO-Rs in a GPA-Dependent Manner in vitro and in vivo.

Diverse forms of Targeted EPO activate EPO-Rs in a GPA-dependent manner in vitro. TF-1 cells, which express both GPA and EPO-Rs,⁵ were treated with several forms of Targeted EPO, and their proliferation was measured 72 h post-treatment (Figure 2a,b). Unfused EPO-(R150A) showed about 100-fold-reduced activity relative to wild-type EPO (EPO(WT)). When EPO(R150A) was fused to an anti-GPA antibody element, its activity was rescued by about 10–100-fold (Figure 2a,b). IH4 and 1C3 were the most potent antibody elements in these tests.

Figure 2.

In vitro and in vivo activities of different forms of Targeted EPO. (a) Typical TF-1 cell proliferation assay comparing the stimulation of proliferation by “wild-type” EPO (including C-terminal tags), the EPO(R150A) mutant protein, and the IH4-35-EPO(R150A) fusion protein. The effective concentration (EC₅₀) was calculated by a four-parameter fit. (b) Table summarizing the ability of different forms of Targeted EPO and control proteins to promote proliferation of TF-1 cells in vitro. (c, d) In vivo test of the ability of IH4-5-EPO(R150A) to specifically promote RBC production and not platelet production using transgenic mice that express human GPA on their RBCs.⁴ The results indicate that IH4-5-EPO(R150A) and diverse other forms of Targeted EPO show enhanced activity of the EPO(R150A) element independent of the mode of attachment of EPO to the GPA-binding antibody element. Data represent mean ± standard error of the mean (SEM).

Three configurations of Targeted EPO were tested to address whether restricting the movement of EPO would affect the EPO activity and potentially cause GPA-bound EPO to preferentially act in cis on targeted cells. The conventional single-chain variable fragment (scFv)–EPO configuration is designed as antibody heavy chain–light chain–EPO (“HLE”). We modified it such that EPO would be between the heavy and light chains; this is feasible because the N- and C-termini of EPO are close in the folded structure. The resulting configurations (“HEL” and “LEH”) were expressed and tested. Both 10F7-EPO(R150A) and 1C3-EPO(R150A) stimulated proliferation of TF-1 cells largely independently of the configuration, indicating that orienting EPO in one way does not significantly constrain its receptor binding in cis (Figure 2b and Table S1).

The GPA-mediated enhancement of EPO(R150A) activity was independent of the linker length in the various fusion proteins. Changing the linker of 10F7-EPO(R150A), R18-EPO(R150A), R18-EPO(K45D), IH4-EPO(R150A), and IH4-EPO(K45D) from 5 to 35 amino acids (about 10–80 Å) in fusion proteins containing the same antibody and EPO elements showed similar EC₅₀ values in cell-based assays (Figure 2b and Table S1). Results with the rotationally constrained and short-linker constructs are consistent with the idea that GPA is an intrinsically disordered, highly flexible protein, so that diverse fusion proteins bound to GPA can still allow binding of EPO with EPO-R.

When tested in vivo, various forms of Targeted EPO specifically stimulated production of RBCs but not platelets. We assessed the in vivo target cell specificity and efficacy of several Targeted EPO proteins in mice that are transgenic for human GPA. Darbepoetin, an extended-half-life form of EPO, was used as a control to represent nontargeted EPO activity. Mice were injected intraperitoneally (ip) with vehicle, darbepoetin, or a Targeted EPO, and their reticulocytes and reticulated platelets were measured to indicate the activities on target and nontarget cells, respectively. Mice treated with darbepoetin showed comparable increases in both new RBCs and new platelets (Figures 2c,d and S2).⁴ Mice treated with IH4-5-EPO(R150A) showed a similar increase in reticulocytes whereas the reticulated platelets were not significantly changed (Figure 2c,d). IH4-5-EPO(R150A) was the most potent of the tested forms of Targeted EPO in vivo (Figures 2b and S2).⁴

Antibody Element IH4 Does Not Induce “RBC Inflammation”.

The anti-GPA antibody element in forms of Targeted EPO might induce an inflammatory phenotype that could enhance blood clotting or cause other undesired side effects and also confound interpretation of experimental treatment with fusion proteins. Various anti-GPA antibodies and V region elements induce a constellation of effects on RBCs, including increased stiffness during shear stress, decreased membrane fluidity, secretion of ATP, production of reactive oxygen species, and phosphorylation of Tyr8 of band 3 (see the Supporting Information). Divalent immunoglobulin G (IgG)-type antibodies against GPA induce at least some and likely all of these phenotypes, but only a subset of monovalent antigen-binding fragments, scFvs, and VHHs induce this RBC response in an epitope-dependent manner (Figure 1e).^10–15

The single-chain camelid element (a VHH or nanobody) IH4 itself lacks stiffening/inflammatory activity when it is bound to GPA and therefore is a good candidate targeting element.¹³ This element binds to the membrane-proximal epitope (₅₂YPPE₅₅) on human GPA.¹⁴ We tested an optimized variant of IH4 (Figure S3). We confirmed by fluorescence recovery after photobleaching (FRAP) that 10F7 induced lower membrane lipid mobility in RBCs whereas IH4 did not (Figure 3). In addition, IH4 appears to bind most tightly of the known anti-GPA V region elements. Therefore, IH4 was chosen as a candidate antibody element for a Targeted EPO that would not induce undesired signaling and would also limit potential interactions of EPO with EPO-Rs on other cells.

Figure 3.

Reduction in membrane fluidity by 10F7 but not IH4, as measured by FRAP. RBC membranes were stained with a lipid-soluble dye, a portion of the membrane was bleached with a laser, and diffusion of the dye into the bleached region was measured. (a) Serial snapshots of RBCs treated with antibody fragments during FRAP experiments. Each recording was made at eight frames per second for 10 s. (b) Mobile lipid fraction of eight or nine RBCs (Experiment 1) or 16–20 RBCs (Experiment 2) treated with buffer, 10F7, or IH4. When the data from these experiments were normalized to the buffer controls in the same experiments and then combined, the difference between 10F7 and buffer (p = 0.0648) or 10F7 and IH4 (p = 0.0463) was significant (Mann–Whitney test). Lines indicate means and 95% confidence intervals. The distribution of individual cell responses is typical for these assays.¹⁶

Geometry of GPA and Targeted EPO Determines Potential for Cell Cross-Linking.

In the design of Targeted EPO, the intention is that the antibody element and EPO bind in cis to receptors on the same cell surface (Figure 1b). However, fusion proteins comprising two binding elements could cross-link different cells bearing a receptor for a component of the fusion protein. To address this, we tested the accessibility of antibodies to the EPO element of Targeted EPO variants bound to RBCs and the ability of different Targeted EPOs to promote binding between RBCs and tumor cells expressing EPO-Rs.

When forms of Targeted EPO are bound to RBCs via GPA, the antibody accessibility of the EPO element decreases when the linker length is shortened and the epitope on GPA is closer to the cell membrane. We compared a series of proteins in which His₆-tagged EPO was fused to IH4 or 10F7 and the linker length was varied. RBCs were incubated with the fusion protein, and then the accessibility of bound EPO was assessed via flow cytometry by binding of a phycoerythrin (PE)-conjugated anti-His₆ antibody. For both IH4 and 10F7 fusion proteins, the PE signal progressively decreased as the linker was shortened (Figure 4a). In addition, cross-linking of RBCs by the divalent anti-His₆ antibody increased with the linker length of the fusion protein, as shown in forward scattering signals evidenced by a shoulder to the right of the main peak (Figure 4b). Regardless of the linker length, IH4 fusion proteins had lower accessibility and fewer RBC cross-linking events by anti-His₆ antibody than 10F7 fusion proteins, suggesting that the binding epitope of the antibody element also affects the accessibility of EPO (Figure 4a,b).

Figure 4.

Targeted EPO fusion proteins with short linkers and membrane-proximal GPA binding epitopes reduce EPO accessibility and cell cross-linking. (a, b) Targeted EPOs containing either 10F7 or IH4 and varying linker lengths were incubated with RBCs and stained with a PE-labeled anti-His₆ antibody. (a) Accessibility of the EPO element by anti-His₆-PE antibody, shown as the mode fluorescence intensity (PE). (b) Cross-linking of RBCs mediated by divalent anti-His₆ antibody, shown as forward scattering (particle size). A signal greater than the cutoff (dotted line) indicates adhesion of RBCs. (c–e) RBCs treated with Targeted EPO variants were incubated with A2780 or MCF-7, both of which express EPO-R: (c) phase-contrast images (200× magnification) of RBC rosetting around EPO-R-bearing cells; (d, e) frequencies of rosette formation, scored as tumor cells bound by three or more RBCs and shown as a percentage of total tumor cells. (f) Relative geometry and distances of GPA and the bound Targeted EPO on the RBC surface. Data represent mean ± SEM.

Forms of Targeted EPO with long linkers and a membrane-distal epitope on GPA mediate adhesion in vitro between RBCs and cells expressing EPO-Rs. We used a rosette assay for cell–cell interaction to address whether the anti-GPA–EPO fusion proteins promote adherence of RBCs to cells from the tumor lines A2780 and MCF-7, which express EPO-R.^17,18 All forms of Targeted EPO containing the 10F7 scFv (epitope at about 34YAATP₃₈) showed a high level of rosette formation, with most or all of the tumor cells binding to three or more RBCs. In contrast, rosette formation with Targeted EPOs containing the IH4 nanobody (epitope: ₅₂YPPE₅₅) was much lower. Targeted EPO forms with shorter linkers generally induced less rosette formation. These effects were synergistic: IH4-EPO(R150A) with a five amino acid linker showed essentially no rosette formation by either A2780 or MCF-7 cells (Figures 4c–e and S4).

The results on antibody accessibility and rosette formation are consistent with a geometric model of the RBC surface in which GPA extends from the cell and shields the cell from interactions with other cells. There are about 800 000 GPA monomers⁶ in a surface area of 140 μm²,¹⁹ such that the average distance between GPA molecules is about 130–190 Å (Figure 4f and S6). The membrane-distal N-terminal ~45 amino acids of GPA include 17 sites for O- and N-glycosylation, with oligosaccharides that extend at least about 20 Å from the peptide backbone (for the O-type blood group antigen).²⁰ It is thought that the N-terminal 45 amino acids of GPA are largely unstructured and would be about 90–130 Å long in an extended conformation.²¹ By comparison, in an scFv–(SGGGS)–EPO fusion protein, the components would be 45, 15, and 51 Å end-to-end (Figure 1f). Thus, contact between a GPA-bound IH4-(SGGGS)–EPO molecule and IgG antibodies or EPO-Rs on another cell should be sterically very difficult (Figure 4f).

IH4-5-EPO(R150A) Does Not Shorten Tail Vein Bleeding Time in Mice.

We tested IH4-5-EPO(R150A) for systemic effects on thrombosis in mice. HuGPA-transgenic mice were injected ip with vehicle, darbepoetin, 10F7-35-EPO(R150A), or IH4-5-EPO(R150A) (Figure 5a). EPO proteins were injected at a dose that resulted in a similar effect in mice as measured by reticulocyte counts. In mice treated with darbepoetin or 10F7-35-EPO(R150A), the bleeding time was shortened by about 30%, whereas mice treated with IH4-5-EPO(R150A) showed no change in bleeding time (Figures 5b,c and S1). These results indicate that IH4-5-EPO(R150A) does not enhance thrombotic side effects, in contrast to the nontargeted form of EPO that acts on EPO-Rs on multiple cell types. When the data from these experiments were normalized to the vehicle controls of the same experiments and then combined, the difference between vehicle and darbepoetin (p = 0.0098) or IH4-5-EPO(R150A) and darbepoetin (p = 0.017) was significant. When data were compared across all experiments with dose groups including 10F7-35-EPO(R150A) and epoetin alfa, these molecules also showed a statistically significant decrease in bleeding time relative to the vehicle control (Figure S1). These results are consistent with previous studies indicating that nontargeted EPO proteins reduce bleeding time in rodents.²²

Figure 5.

Shortened tail bleeding time in mice treated with a nontargeted EPO but not those treated with IH4-5-EPO(R150A). (a) Experimental approach. Mice received a single ip injection of darbepoetin (50 pmol; 1.8 μg), IH4-5-EPO(R150A) (40 pmol; 2 μg), or vehicle. After 24 h, tail transection was performed, and bleeding times were recorded (see Methods). (b) Three experiments showing bleeding times relative to the vehicle control in huGPA-transgenic mice treated with vehicle, darbepoetin, or IH4-5-EPO(R150A). Data represent median with interquartile range. (c) Kaplan–Meier plot for Experiment 1 from the data shown in (b). Some images were created with BioRender.com.

CONCLUSIONS

The design of a fusion protein must consider the spatial orientation of the target proteins to function while minimizing unwanted interactions. Such considerations become important in AND-gate bifunctional protein systems that bind to different receptors abundant on different cell types. EPO-Rs are found on numerous cell types besides RBC precursors, including platelet precursors, vascular endothelial cells, liver cells, and white blood cells such as macrophages.

Here we addressed this problem for “Targeted EPO” molecules, fusion proteins that bind to GPA and weakly to EPO-R and thus are AND-gated to bind only to cells with both of these transmembrane proteins—namely, late RBC precursors. The design goal is to activate EPO-R only on late RBC precursors as a result of initial binding to GPA, followed by binding in cis to EPO-R due to the high local concentration of EPO (Figure 1b). We found the following:

1. Some Targeted EPO proteins mediate adhesion between mature RBCs that express only GPA and other cells that express only EPO-R.

2. The position of the epitope on GPA and the length of the linker between the GPA binding element and the EPO element determine whether a Targeted EPO mediates cell adhesion.

3. The cell adhesion results can be interpreted with a structural model in which GPA extends from the RBC surface and limits how close another cell can approach. To mediate cell adhesion, the EPO element must extend beyond the membrane-distal end of GPA.

4. More broadly, cell–cell interactions may be regulated by abundant cell-surface proteins that may sterically prevent close approach of cells. This needs to be considered when constructing engineered cell interaction systems, such as CAR-T cells and bispecific antibodies.

A major challenge for synthetic biology is that artificial biological systems have the potential for undesired interactions that can lead to system failure. When we study natural systems, we are generally not aware of all the behaviors they could have if they were poorly designed. Our experiments illustrate geometric principles that natural systems may use in achieving specificity.

METHODS

Methods for cell culture, DNA construction, protein expression and purification, in vitro measurement of EPO activity by cell proliferation assays, measurement of mouse reticulocytes and reticulated platelets, and FRAP were performed as described previously.^4,16 See Supplementary Methods.

EPO Accessibility and RBC Cross-Linking Measurement by Flow Cytometry.

RBCs (Zen-Bio, Research Triangle Park, NC; 0.2% v/v in PBS-glucose) were incubated with fusion proteins for 1 h at 4 °C, washed once with PBS-glucose (PBS, 5 mM glucose, pH 7.4), incubated with anti-His₆-PE antibody (Abcam) for 30 min at 4 °C, and washed once with PBS-glucose. The fusion protein concentration used was calculated on the basis of the K_D values for the monovalent 10F7 and IH4 antibody elements (95 and 33 nM, respectively) to achieve 80% saturation of GPA. The cells were analyzed by flow cytometry. See Supplementary Methods.

RBC Rosette Assay.

A2780 and MCF-7 cells were plated on a six-well plate containing 12 mm × 12 mm coverslips and incubated overnight at 37 °C. On the next day, a mixture of RBCs (0.2% v/v) and fusion proteins was prepared in RPMI-1640 medium without bicarbonate and incubated for 30 min at 23 °C. Coverslips with cells were washed twice with PBS and transferred to a new six-well plate. The RBC/fusion protein mixture was added on top of each coverslip and incubated for 1 h at 37 °C. Coverslips were gently washed twice with PBS. Rosettes were imaged on a Nikon Eclipse TE300 inverted phase-contrast microscope. Rosettes were defined as adherent cells bound by three or more RBCs among at least 160 cells from at least eight fields. See Supplementary Methods for further details.

Mouse Tail Vein Bleeding Time Measurement.

All of the mouse experiments were performed under the protocol IS00000723, approved by the Harvard Medical School IACUC. Experiments were performed in a blinded manner. Ten mice per dose group received a single ip injection with saline or test protein. On the next day, mice were anesthetized. The tail was transected 3 mm from the tip, and the body of the mouse was placed on a heated pad over a 37 °C water bath containing tubes with 50 mL of saline, such that 0.5–2 mm of the transected tail was placed in the saline within 2 s of transection. The bleeding time was recorded when bleeding stopped for 1 min. Bleeding times longer than 10 min were recorded as 10 min. See Supplementary Methods for detailed a description.

Supplementary Material

ASSOCIATED CONTENT

Supporting Information

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acssynbio.9b00273.

Additional description of methods, interpretation of data, and explanation for the structural model; tables and figures of cumulative in vitro and in vivo data; a list of anti-GPA antibodies and their properties; lists of DNA and protein sequences used in this study (PDF)