Engineering Agonistic Bispecifics to Investigate the Influence of Distance on Surface-Mediated Complement Activation

This article is featured in Top Reads, p. 107

Inducing specific Ab effector functions to activate the human innate immune system has become a therapeutic method to treat diseases (1). Ab binding to cell-surface Ags leads to Ab-dependent cellular phagocytosis, Ab-dependent cellular cytotoxicity, and complement-dependent cytotoxicity (CDC). CDC relies on activation of the classical complement pathway by Ag-bound IgG or IgM complexes, which form a hexameric platform via interactions between the Ab Fc domains (2–5). This hexameric platform binds the first complement component, C1, which comprises a heterotetramer of two proteases, C1r and C1s, and the pattern recognition receptor C1q (6). Binding of the globular head domains of C1q (gC1q) to hexameric Ab platforms induces activation of C1r and C1s. C1s proceeds to cleave complement protein C4 to form C4b, an opsonin that binds covalently to surrounding molecules. C2 binds to C4b, before C1s also cleaves C4b-bound C2, forming the C4b2b complex, known as a C3 convertase, which propagates the classical complement pathway leading to further opsonization. Under ideal conditions, complement terminates with formation of the membrane attack complex (MAC) pore and lysis of the targeted cell (7).

Complement can be activated by Ag-bound IgG1, IgG2, IgG3 or IgM Abs (8, 9). Recent structures of C1 bound to IgG1 (5), IgG3 (4), and IgM (3) have revealed structural differences during complement activation. Most notably, hexameric IgG3 is significantly taller than IgG1 and IgM, at 22, 11, and 13 nm, respectively. IgG3 is also the most potent IgG subclass regarding CDC (8, 10–13), implying that increasing height can enhance complement activation. However, this correlation may also be affected by other parameters, including the ability of IgG3 to undergo divalent Fab binding (4), whereas IgG1 must bind monovalently for effective complement activation (2); the extended hinge unique to IgG3 renders this subclass more flexible than others (14); and C1-bound IgG3 deposited C4b directly onto IgG3, whereas both IgG1 and IgM deposited C4b onto surrounding molecules. How these structural and biochemical differences affect CDC remain unknown.

CDC has previously been seen as an adjunct for Ab-mediated therapy. However, C1 has recently become a therapeutic target itself (15–17), as CDC has been shown to be the primary effector function of several immunotherapeutics (18). This is exemplified by recent Ab engineering successes that specifically target and subsequently activate C1 (15–17, 19, 20). In particular, bispecific Abs and nanobodies (Nbs) have been used to recruit C1 to both mammalian cells and bacteria (17, 20, 21).

Nbs are single-chain Ag-binding domains comprising only the Ag-binding fragment of H chain–only camelid Abs (22), and are only one-tenth the size of an IgG1 Ab (15 versus 150 kDa, respectively). Whereas Abs comprise both targeting (Fab) and effectuating (Fc) domains, Nbs comprise only targeting moieties. Nbs therefore lack the Fc domain of Abs and are consequently unable to form hexameric platforms analogous to IgG1, IgG3, and IgM for CDC. In keeping with this, a recent C1q-binding Nb inhibits complement activation by blocking Ab binding (23). However, when targeted to membranes, Nbs that bind to C1q have been shown to activate the C1 complex (20). These data indicate that the same C1-targeting Nb can act as an antagonist in solution and as an agonist when membrane bound. To target the C1 complex to specific cell types, bispecific Nbs composed of two distinct genetically fused Nb monomers are used, where one Nb domain binds C1q while the other Nb domain binds to the target molecule. Direct engagement of C1 by bispecific Nbs may therefore represent a new mechanism for complement activation, and it is actively being pursued as a therapeutic route (21).

C1q-binding Nbs also represent ideal tools to explore C1 activation in minimal, synthetic systems; Nbs contain only the targeting domain, and compared with Abs are more genetically tractable and easier to engineer. Furthermore, Abs have a propensity to form hexameric platforms via noncovalent association of Fc domains (24), which limits the ability to control Ab oligomerization. Using monomeric Nbs is therefore a more controllable method to determine the monovalent binding requirements of complement activation. In this study, we use bispecific Nbs to position C1q distinct distances away from lipid membranes to explore the height requirements of complement activation. We perform this using both DNA nanotechnology and protein engineering, and reveal the effect of linker length and flexibility between Ag and C1 binding. Such knowledge explains why certain Ags do or do not work, and furthermore could contribute to choosing and selecting targets for bispecific constructs able to activate complement.

Genes were ordered from GeneArt and codons were optimized for Escherichia coli expression systems using the GeneArt software. All enzymes and kits used for cloning were purchased from New England Biolabs and used according to the manufacturer’s instructions. The protein constructs were amplified using Q5 polymerase, purified using PCR clean-up kits and then treated with restriction enzymes NcoI-HF and HindIII-HF. Ligation was performed using T4 ligase to insert constructs into pCPF3.05 plasmids and the product was transformed into DH5α competent cells. All plasmids were sequenced using Sanger sequencing and sequencing was performed by the Leiden Genome Technology Centre.

Protein expression was performed using BL21 DE3 competent cells in lysogeny broth. All cell cultures were oxygenated via shaking at 200 rpm. Cells were grown at 37°C until an OD₆₀₀ of 0.6 was reached and expression was induced using 0.5 mM isopropyl β-d-1-thiogalactopyranoside (IPTG). Cells were then transferred to a 20°C incubator for 16 h. After overnight incubation cells were collected by pelleting and resuspended in cold wash buffer containing 300 mM NaCl, 20 mM Tris-HCl, and 20 mM imidazole (pH 8) (wash buffer). Lysis was induced using probe sonication, and debris was removed from solution by centrifugation at 24,000 × g for 40 min at 4°C. Subsequently HisPur Ni-NTA resin was equilibrated with 10 column volumes of wash buffer. Samples were loaded onto the column and washed with 10 column volumes of wash buffer and then eluted using 250 mM imidazole, 300 mM NaCl, 20 mM, and Tris-HCl (pH 8) elution buffer. Proteins were then purified further using size exclusion chromatography. Samples were loaded onto a S200 Superdex prep grade column that was equilibrated with PBS on a Bio-Rad NGC chromatography system. Samples were collected and concentrated using an Amicon Ultra-15 centrifugal filter and stored at −80°C. Purity was analyzed via SDS-PAGE and mass spectrometry. For mass spectrometry analysis a UPLC protein C4 column (2.1 × 50 mm) was used on a gradient of eluent as follows: A, water + 0.1% formic acid; and B, acetonitrile + 0.1% formic acid. The gradient ran from 2% B to 100% B for 1.6 min at 60°C. After the liquid chromatography separation the flow though was analyzed on a Waters Xevo G2-XS QTof mass spectrometer using electrospray ionization, and mass data were measured from 50 to 2000 m/z.

The alfaTag peptides used in the experiments presented in this study contain either a cholesteryl hemisuccinate or an azide moiety on the N terminus. The alfaTag, with sequence SRLEEELRRRLTE, was produced using solid-phase peptide synthesis on a Liberty Blue microwave-assisted peptide synthesizer. A rink-amide resin was used as the solid support, and synthesis was performed using standard Fmoc chemistry methods. A 20% piperidine in dimethylformamide (DMF) mix was used to remove the Fmoc groups, and Oxyma/DIC were employed as base/activator, respectively, for the coupling steps. After synthesis, the resin was washed using DMF and the N terminus of the peptide was functionalized using either 6-azido-hexanoic acid or cholesteryl hemisuccinate in a DIPEA/HCTU mix, all at three equivalents as compared with the peptide. This reaction was left overnight and the resin was subsequently washed three times using DMF and then dichloromethane. Peptides were cleaved off the resin using trifluoroacetic acid (TFA) with 2.5% triisopropyl silane and precipitated in a 50/50 (v/v) mix of ice-cold diethyl ether/hexane. The precipitate was redissolved in 25% MeCN in water and freeze-dried. Crude peptides were purified using reversed-phase liquid chromatography. The cholesteryl-modified peptide was purified on a C4 column, and a C18 column was employed for the azide functionalized peptide. A gradient of 20–80% MeCN (containing 0.1% TFA) was applied, with the other solution being ultrapure water (containing 0.1% TFA). Fractions were then collected and analyzed using liquid chromatography–mass spectrometry, pooled, and freeze-dried. Dry peptide was stored at −20°C until use.

dsDNA constructs were developed using the NANEV software (25), and predictions on folding and the orientation of each functional group was verified using NuPACK software (Ref. 26 and M.E. Fornace, J. Huang, C.T. Newman, N.J. Porubsky, M.B. Pierce, and N.A. Pierce, manuscript posted on ChemRxiv, DOI: 10.26434/chemrxiv-2022-xv98l). DNA strands were tabulated in Supplemental Table I.

All custom DNA strands were ordered from Integrated DNA Technologies. Biotin and cholesterol modifications were purchased commercially, and the alfaTag modification was covalently linked to the DNA via an amine modification on the DNA. The DNA-amine was linked to a heterobifunctional NHS-DBCO linker (dibenzocyclooctyne-N-hydroxysuccinimidyl ester, Merck). For the NHS amine coupling a 50:1 linker/DNA molar excess in PBS containing 50% (v/v) DMSO was left at 37°C and 1000 rpm on a ThermoMixer F2.0 (Eppendorf, Hamburg, Germany) for 12–16 h. The excess linker was removed using 10-kDa Amicon spin columns. Subsequently, azide functionalized alfaTag, which was produced in-house (see Peptide synthesis), was added in a 5:1 peptide/DNA ratio in PBS containing 50% (v/v) DMSO. This mixture was left at 25°C and 1000 rpm for 12–16 h. The conjugated product was purified using ion-pairing HPLC using a C18-coated stationary phase. For the ion-pairing agent, 0.1 M triethylammonium acetate (Sigma-Aldrich, St. Louis, MO) was used and set to an acetonitrile gradient (CAN; VWR, Radnor, PA). The purified product was freeze-dried overnight and resuspended in nuclease-free water and stored at 4°C.

Strands were folded together in stoichiometric ratios. Before handling cholesterol functionalized DNA handles, these were incubated at 50°C for 5 min to disrupt cholesterol aggregation. All folding reactions were performed in 20 mM tris (pH 8), 5 mM NaCl, and 25 mM MgCl₂. Annealing of the monomers was performed at 1200 nM concentration. The DNA monomers were annealed from 80 to 70°C at −0.2°C every 1 min, from 70 to 30°C at −0.1°C every 1.30 min, and from 30 to 20°C at −0.1°C every minute and kept at 20°C for 2 h. Annealing was carried out using a Bio-Rad C1000 Touch thermal cycler. The monomeric constructs were incorporated into the liposomes immediately after folding. To assess folding and size of the DNA constructs, 10% polyacrylamide gels were made using TEA buffer, run for 1.5 h at 100 V on a Bio-Rad Mini-PROTEAN system, and stained using GelRed. For gel electrophoresis analysis, DNA was folded in the absence of functional groups, as these impaired size analysis. Controls for incorporation of functional groups were performed during functional assays.

ELISAs were performed on Maxisorp Nunc immunoplates (Thermo Fisher Scientific, Waltham, MA). For the complement deposition analysis, 10 µg/ml streptavidin was added to the well plate in 0.1 M Na₂CO₃, 0.1 M NaHCO₃ (pH 9.6) at 50 µl per well and allowed to incubate either at room temperature overnight or 37°C for 1 h. After incubation the plate was washed three times PBS with 0.05% Tween 20 (v/v). For each following step the plate was incubated at 37°C for 1 h and subsequently washed three times in PBS with 0.05% Tween 20. Wells were blocked with 100 µl of 0.1 M spermidine in ultrapure water (milliQ). After blocking biotin–DNA–alfaTag, constructs and bispecific proteins were titrated, incubated, washed, and subsequently incubated with 1% normal human serum (NHS) (Complement Technology) in RPMI 1640 medium. Detection of either His-tag, C1q, C4, or C5–9 was performed using a primary and secondary Ab pair. Primary mouse anti–6× His-tag (1:2,000) (Thermo Fisher Scientific), rabbit anti-C1q (1:2,000) (Dako), goat anti-human C4 (1:150,000) (Complement Technology), or mouse anti-human 5b–9 (Dako) was added to PBS, 0.05% Tween 20, 1% BSA (PBS-BT). Then, donkey anti-goat HRP (Invitrogen), goat anti-rabbit HRP (Dako), and goat anti-mouse HRP (Dako) Abs were diluted 1:5000 in PBS-BT and added to the plates. To detect the secondary Ab, HRP with 2.5 mg/ml ABTS added in citric acid buffer (0.15 M, pH 4.2) (ABTS buffer) containing 0.15 (v/v %) H₂O₂ was used. Absorption measurements were performed on a CLARIOstar microplate reader (BMG Labtech, Offenburg, Germany). One variation on this protocol was made when comparing several dilutions of NHS and to determine C1q binding capacity to bispecific camelid (BC) variants. In this study, BCs were coated to the plate in the same buffer and concentration as streptavidin. Wells were coated with spermidine, after which NHS was immediately added. Detection was performed as described.

Lipids were prepared in the following molar ratios: 1,2-dimyristoyl-sn-glycero-3-phosphocholine/1,2-dimyristoyl-sn-glycero-3-phospho-(1′-rac-glycerol) (sodium salt)/ cholesterol (ovine) (DMPC/DMPC/CHOL) 45:5:50. In the conditions where the cholesterol/alfaTag was incorporated, the ratios were modified to DMPC/DMPC/CHOL/alfaTag 44.5:5:50:0.5. All lipids except for the custom-made cholesterol/alfaTag were ordered from Avanti Polar Lipids. Lipids were dissolved in a chloroform/methanol mixture 9/1 (v/v) in glass vials. Lipids were added in the appropriate ratios up to 1 mg of total lipid mass, and the chloroform methanol mix was allowed to evaporate under a stream on N₂ for at least 2 h or overnight. Then, liposomes were hydrated using aqueous buffer. The hydrated lipid films were detached from the glass vials using a water bath at 50–60°C and vortexing. Liposomes were formed using sonication at 50–60°C for 10 min in a Branson 2800 ultrasonic cleaner. These liposomes were stored at 4°C until further use and were stable until at least 1 mo after synthesis. For the DNA monomer experiments, the incorporation of the DNA into the lipid bilayer was achieved after liposome formation. Cholesterol-modified DNA was coincubated with liposomes at 25°C and 500 rpm on a tabletop shaker for 1 h, after which liposomes were centrifuged at 20,000 × g for 15 min at 4°C, supernatant was removed, and liposomes were resuspended in buffer.

Liposomes containing the cholesterol-modified DNA-alfaTag were incubated with BC16 and 1.5% NHS at room temperature for 30 min. Aliquots were taken and mixed 1:1 with Laemmli buffer (Bio-Rad), and DTT was added to a concentration of 50 mM. Samples were incubated at 95°C for 5 min before being added to 4–12% (Bis-Tris) PAGE gel (mPAGE). Blotting was performed in 7.2 g/l glycine, 14.5 g/l Tris in MQ water with 10 vol% methanol. Temperature was kept low using iced water, and transfer was performed at 10 W for 3 h. Blots were washed using PBS with 0.1% Tween 20 (PBST) and blocked using PBST with 5 mass/vol% low-fat powdered milk (Biologische Magere Melkpoeder; Koe-Vache; Mattisson Healthstyle). C4 was detected using goat anti-human C4 (Complement Technology, A205) 1:150,000 dilution and donkey anti-goat HRP (Invitrogen, PA1-28664). Samples were visualized using Clarity Western ECL substrate (Bio-Rad) according to the manufacturer’s instructions and imaged in a ChemiDoc imaging system (Bio-Rad).

Liposomes containing cholesterol-modified DNA-alfaTag in 150 mM NaCl, 50 mM Tris-HCl, 5 mM CaCl₂, and 10 mM MgCl₂ (pH 7.5) (assay buffer) containing 500 µM Boc-Leu-Gly-Arg–amino methyl coumarin (LGR-AMC) (PeptaNova, Sandhausen, Germany), diluted from a 10 mM stock (5% DMSO final concentration), were used for this assay. Liposomes were formed or resuspended in assay buffer and formed as described under Liposome preparation. C1 protein (Complement Technology) was kept at 40 nM final concentration. C1 purified protein was buffer exchanged to assay buffer using 100-kDa spin filters. C1s substrate (LGR-AMC) was monitored over time on a CLARIOstar microplate reader (BMG Labtech, Offenburg, Germany). The enzyme conversion was measured without any further purification methods, all components were added in a single reaction, and C1s substrate conversion was measured over time. Measurements were taken every minute for a period of 5 h with excitation and emission set at 360 and 460 nm, respectively. After data collection, fluorescence intensity over time was transformed to AMC conversion per minute. To obtain the conversion rate a linear plot was fitted to the data generated between 200 and 300 min of incubation. The standard curve of AMC to fluorescence was generated using three separate 10 mM AMC stock solutions (DMSO) titrated into assay buffer with DMSO kept at 5% final concentrations. The model was fitted using GraphPad Prism (version 9.3.1).

Liposomes were prepared to contain 20 mM sulforhodamine B (SRB) (Sigma-Aldrich, St. Louis, MO). Samples were preincubated for 2–5 min with 10% NHS and 50/50 vol% PBS and RPMI 1640 medium before adding bispecifics to activate complement. For every burst assay, liposomes were monitored for at least 2 min on the plate reader to assess their stability prior to data acquisition. The liposome release of SRB was monitored over time (every 15 s) on a CLARIOstar microplate reader (BMG Labtech, Offenburg, Germany).

Graphs of data were made and statistical analysis was performed using GraphPad Prism (version 9.3.1). Cartoons and visual representations of proteins and DNA were made using ChimeraX (version 1.5) (27).

To recruit C1, we used an existing Nb, C1qNB75, that binds with high affinity to C1q (K_D of 300 pM, compared with ∼100 µM for the interaction of a single gC1q domain with monomeric IgG) (28), and at the same location on gC1q as Abs (23). This Nb has been previously shown to activate complement on cells (20, 21). To target C1qNB75 to surfaces, we created a bispecific Nb by forming a tandem dimer with another Nb called NBalfa, which binds strongly (K_D of 27 pM) to a synthetic Ag comprising 13 aa that adopts a compact α helix, known as an alfaTag (29). The alfaTag was synthesized to display a cholesterol moiety at the N terminus, which can be incorporated into lipid membranes, thereby generating model antigenic surfaces (Fig. 1A).

Initially, a BC Nb comprising C1qNB75 and NBalfa linked using a flexible 16-aa linker (BC16, see Supplemental Table II) was expressed and purified (Supplemental Fig. 1). The capacity of BC16 to bind both C1q and the alfaTag was assessed using an ELISA (Fig. 1B). After both Nb domains within BC16 were confirmed to be functional, the capacity of BC16 to activate the C1s protease was determined. Activation of C1s was assessed using a fluorescence assay, whereby a nonfluorescent substrate (LGR-AMC) becomes fluorescent upon cleavage by C1s (30), which can be used to calculate the LGR-AMC turnover rate (31). Liposomes were produced that contained 0.5 mol% cholesterol-modified alfaTag, creating cell-mimetic Ag-presenting membranes. BC16 binds to these liposomes and recruits the purified C1 complex, leading to upregulation of C1s baseline enzyme activity (Fig. 1C). Binding of BC16 to C1 alone does not result in upregulation of C1s activity, indicating that BC16 only has an agonistic effect when bound to the membrane.

The complement pathway terminates with the formation of the MAC pore, which perforates lipid membranes. A self-quenching concentration of fluorescent SRB dye was encapsulated in liposomes during formulation. Upon activation of complement and formation of the MAC pore, the dye is diluted and fluoresces (32). Liposomes containing SRB and displaying alfaTag Ags were incubated with BC16 and human serum, which contains all required complement components of the classical pathway. We observed a rapid increase in fluorescence, indicating successful activation of the classical complement cascade by Ag-bound BC16 (Fig. 1D), with an EC₅₀ of ∼20 nM. This concentration and the rate of activation are highly comparable to those of Abs (3, 32, 33).

Inspired by the height variability of IgG1, IgG3, and IgM Ab Fc platforms (2–5), we explored the importance of linker length, and therefore indirectly distance to membrane, for efficient complement activation by the C1 complex. DNA spacers of various lengths were designed to bind to surfaces at the 3′ terminus, via either a cholesterol moiety (to bind to lipid membranes) or a biotin molecule (to bind to streptavidin) attached during synthesis (Supplemental Table I). By including a primary amine at a specific position during DNA synthesis, a heterobifunctional linker was used to covalently attach the alfaTag peptide to the DNA at the terminus distal to the cholesterol or biotin (Supplemental Table I). This resulted in a series of DNA linkers separating a surface-binding motif and the alfaTag Ag (Fig. 2A).

The linkers comprised a ssDNA oligonucleotide 19 nt long, or dsDNA helices 19, 44, 61, 78, and 95 bp long. These DNA linkers therefore have approximate contour lengths of 6 nm (19-nt ssDNA), 6 nm (19-bp dsDNA), 13, 18, 24, and 29 nm, respectively (Fig. 2A) (27, 34, 35). DNA constructs were thermally annealed at stoichiometric ratios, and the final length of the designs was verified using PAGE.

Cholesterol-modified DNA linkers were incorporated into preformed bare liposomes (which did not display Ags) before the BC16 and purified C1 complex was added and activation of the C1s protease determined using the LGR-AMC substrate described above. Each construct was able to activate C1 significantly more than the negative controls, which were either BC16 alone (with no DNA) or without BC16 added (Fig. 2B). Upregulation of C1s activity was consistent for all DNA strands and was significantly higher compared with the negative controls. Although upregulation was observed for all constructs, the 44- and 95-bp DNA strands were significantly higher and lower than the other constructs, respectively. The poorer activation by the 95-bp construct is in line with the results described below, although the higher level of upregulation of C1s activity by the 44-bp construct did not result in more efficient overall complement activation.

Next, we determined whether these constructs were able to induce MAC pore formation on liposome membranes. Cholesterol-modified DNA linkers were bound to bare liposomes that encapsulated SRB, before BC16 was added. Next, human serum was added and the fluorescence measured. In contrast to C1s activation, not all of the linkers resulted in MAC pore formation (Fig. 2C). The 19-nt ssDNA and 19-bp dsDNA linkers both led to MAC pore formation with similar efficacies. The 44-bp dsDNA linker, although inducing MAC pore formation, did so at slightly lower levels than the shorter linkers. Above this linker length, the ability of the constructs to induce MAC pore formation decreased significantly.

To identify the reason that MAC pore formation decreased with height despite C1s activation remaining high, we performed Western blotting and ELISAs to determine at which point of the classical complement pathway there was a deviation in activation.

The DNA-coated liposomes described above were incubated with BC16 and human serum for 30 min before reduction and denaturation of the sample and separation using SDS-PAGE. Western blotting was then performed to detect C4 components. C4 comprises three disulfide-linked chains, that is, C4α, C4β, and C4γ. Cleavage by C1s results in removal of C4a from the C4α chain to form the shorter C4α′ chain–containing C4b molecule. Western blotting showed equivalent formation of C4α′ in the presence of all necessary components (Fig. 3A), indicating that each of the DNA spacers is equally able to activate the C1 complex and cleave C4.

Next, complement progression was assessed using ELISAs. DNA constructs containing both an alfaTag and biotin (Supplemental Table I) were immobilized on streptavidin-coated ELISA plates. These were then incubated with BC16 and 1% human serum in RPMI 1640 medium. The ability of each construct to recruit C1q in this format did not differ (Fig. 3B). However, the ability to fix C4 was different; longer linkers were less able to deposit C4b (Fig. 3C), even though Western blotting indicated that C1s was equally able to cleave C4. This trend continued with MAC formation, as assessed by detecting the MAC components C5b–9 (Fig. 3D); longer linkers resulted in reduced MAC pore formation. Taken together, these data indicate that the DNA spacers are equally able to bind and activate C1, and that activated C1 is also able to cleave C4. However, as the distance to the membrane increases, cleaved C4b is less able to bind to the surface, and consequently less able to form C3 convertases and advance the complement pathway to MAC pore formation.

Next, we synthesized several bispecific Nbs with different linker lengths separating the C1qNB75 and NBalfa Nbs. This was performed to 1) assess whether these behaved similarly to the DNA-based linkers, 2) move toward a more native, protein-based system, which may have therapeutic utility. These contain the same Nbs as BC16 above, C1qNB75 and NBalfa, but differ in their linker properties; these were designed to assess how the height from the membrane and linker flexibility affects complement activation, as IgG3 contains a long, flexible hinge (4). As well as BC16, these constructs comprised a longer and flexible 60-aa linker (BC60), a long semiflexible linker composed of two tandem-dimer α helices separated by a flexible hinge (BCtdα), and a rigid α helix derived from myosin (BCmyo) (36) (Fig. 4A, Supplemental Table II). For the flexible linkers, BC16 and BC60, the average end-to-end distances were determined as described in Evers et al. (37), plus an additional 3 nm per Nb, whereas the end-to-end distances in the more rigid BCtdα and BCmyo were measured from the atomic models comprising rigid α helices (Fig. 4A). Estimates of the distance that the bispecific constructs are able to bridge are therefore 8.5, 11, 17, and 17 nm for the BC16, BC60, BCtdα and BCmyo, respectively (Fig. 4A) (36–39).

These four separate constructs were produced using bacterial expression and characterized via SDS-PAGE and mass spectrometry (Supplemental Fig. 1). The constructs were compared for their efficacy to bind alfaTag Ags on liposomes and activate complement. Detection of either C1qNB75 and NBalfa using ELISA revealed that binding to alfaTag and C1q was equivalent between the constructs (Fig. 4B, Supplemental Fig. 2A). Subsequently, functional assays were performed to compare complement activation in human serum via liposome lysis by the MAC pore. Each of the constructs was able to activate complement (Fig. 4C, Supplemental Fig. 2B). However, the longest and most rigid linker, BCmyo, demonstrated reduced complement activation compared with the other constructs, in agreement with the DNA spacer data above.

To further test the notion that taller constructs have a reduced ability to activate complement, each of the BC constructs was moved farther from the membrane by combining them with the DNA-based linkers. Liposomes were functionalized with cholesterol-alfaTag, or with cholesterol-modified DNA linkers with spacers comprising 5 nt of ssDNA, 19 bp of dsDNA, or 44 bp of dsDNA. All construct and linker pairs were able to activate complement, but the additional height of the Ag provided by the extended DNA linkers reduced the efficacy of MAC pore formation most for longer linkers, in the order BC16>BC60>BCtdα>BCmyo (Fig. 4D, Supplemental Fig. 2C).

Developing agonists able to bind C1 and activate complement is a novel route to perform targeted cell killing. Bispecific Nbs able to recruit C1 to targets have been shown to successfully activate complement and induce complement-dependent cell killing (CDC) on clinically relevant targets (20, 21). However, inducing CDC is not straightforward, and effectiveness can vary significantly between bispecific designs (20, 21), Ab subclass (13), and epitope (40). This variability may have a structural cause and may also be related to the reason that certain Ab-based immunotherapeutics are better or worse at activating complement (40). In this study, we investigated complement activation using monomeric agonists of C1 in the form of bispecific Nbs, which allowed us to gain control over certain geometric parameters known to be important for C1 activation.

A bispecific Nb capable of binding simultaneously to both a synthetic peptide-based hapten and C1 was characterized (Fig. 1). This initial BC contained a 16-aa linker (BC16) that was similar to work previously published on cellular targets (20, 21), and this enabled comparison with the synthetic systems described in the current study. BC16 only upregulated C1s activity when bound to antigenic surfaces, and no fluid phase activation of the bispecific was observed, which is a requirement if such bispecific Nbs will be developed as a pharmaceutical. This surface-specific activation indicates that multivalent interactions between the surface and gC1q impose a structural constraint on C1, leading to activation, as previously hypothesized (4, 5). Whether this is a compaction of the C1q collagenous arms (5), a structural rearrangement of the C1rs protease platform (3, 4), or a different mechanism remains to be determined. However, the systems described in the current study provide nascent molecules and methods to probe this activation mechanism further.

In this study, we focused on the impact of linker length as an indirect but comparable metric for Ag and epitope location. It is still unclear how the Ab-binding location affects complement activation. This has gained pertinence now that the recent structure of IgG3 displays a much taller Fc platform than IgG1 or IgM (4). Geometry of the bispecifics was investigated by the inclusion of a linker to vary the end-to-end distance of the C1-binding site to the membrane. The contour length of the agonists was iteratively increased, thereby enlarging the radius of gyration from the anchor point of the membrane. End-to-end distances between 5 and 18 nm were broadly similar (Fig. 2), but activation was reduced when C1 was located beyond 18 nm from the surface. This value was consistent for both the DNA-based and protein-based linkers. This was expected, as both of these systems mediate the positioning of C1qNB75, the Nb that binds to C1, but the finding is nevertheless important from a synthetic biology view, as it shows that DNA nanotechnology can be used to modify the geometric parameters of complement activation (41).

To determine the cause of this height-induced difference, we assessed various stages of complement activation, including C1q binding, C1s activation, C4 cleavage, C4b deposition, and terminal pathway activation (Figs. 2, 3). Although all constructs were equally effective at cleaving C4 (Fig. 3A), they differed in their ability to deposit C4b (Fig. 3C); increasing linker length led to more ineffective C4b deposition, which subsequently impaired MAC pore formation. A discrepancy in C4b deposition has also been observed in native ligands for C1 (42). In particular, IgG isotypes have been observed to display discrepancies in C1q binding and efficiency of C4 deposition (42). Although IgG1 and IgM are only 11 and 13 nm tall, respectively (3, 5), IgG3 allotype G3m5 has a hinge length of 62 aa, resulting in an Fc platform 22 nm above the Ag (4). While IgG1 and IgM are short enough for C1 to deposit C4b directly adjacent to the Ab platform, IgG3 instead deposits C4b onto the Fab domains of the Abs themselves, presumably due to the distance between C1s and the membrane being too large. These structural data, combined with the observations from this study, led to the hypothesis that C4b deposition onto membranes is limited by the size of the C1 ligand and exceeding this limit leads to impaired formation of the MAC, and consequently CDC. An additional parameter is the presence of two isotypes of C4 in human serum, that is, C4A and C4B. Although both are able to bind to the liposome surfaces used in the current study, C4A and C4B have different hydrolysis rates and preferential chemical targets, with C4A and C4B reacting preferentially with amino and hydroxyl groups, respectively. The liposomes used in this work contain hydroxyl groups and as such will preferentially bind C4Bb (43). Although the liposome composition is consistent between experiments, it will be intriguing to discover how the C4 isotypes are affected by the distance between the sites of C4 cleavage and deposition.

Taken together, these data strongly imply that the epitope location affects complement activation by reducing the amount of C4b that is fixed, but will also be affected by the Ab class and subclass. Indeed, previous research has implied this (44, 45); increasing the distance between surface and epitope led to reduced CDC (45), while IgG3 allotypes display decreasing CDC with increasing hinge length (44). In this study, we explain these findings with the notion that this difference is due to impaired C4b deposition, and not due to C1 binding, activation, or C4 cleavage.

As well as the overall distance from the membrane to the C1 complex, the degree of linker flexibility can further determine the efficiency of the complement activation. Although MAC pore formation via BCmyo plateaued at a significantly lower level as compared with the other BC variants (Fig. 4), the BCtdα and BCmyo protein linkers are expected to be able to bridge similar distances (36, 46). The semiflexible linker of BCtdα, however, which comprises two ∼5=nm-long tandem α helices connected by a flexible linker, is expected to be more disordered compared with the ∼10-nm helix of the BCmyo construct. This indicates that impeding the ability of the C1 complex to reach the surface, as for the rigid BCmyo construct, is also detrimental to C4b deposition and CDC.

CDC can be caused by IgG1, IgG2, IgG3, or IgM Ab classes. However, in native serum, Abs exist as polyclonal populations, and therefore bind to the same Ag at different epitopes. How these Abs interact to form an Fc-mediated oligomeric surface required for complement activation is unknown. The different heights of IgG1 and IgG3 are therefore but a snapshot into the structural requirements of C1 binding and activation, and the impacts of linker lengths, epitope location, and Ab class and subclass on complement activation remain to be determined. Furthermore, other Ab-mediated processes, such as Ab-dependent cellular phagocytosis, are known to be influenced by Ag height differences (47). In this study, we have generated and characterized proteins and constructs that not only provide insights into these mechanisms, but also support the notion that linker and hinge engineering could present an attractive method to increase or gain potency over a certain cellular target.

The authors have no financial conflicts of interest.

We thank Prianka Luther for guidance during the synthesis of the alfaTag peptides, and Bjorn Doodewaerd for help with mass spectrometry.