Improved GPCR ligands from nanobody tethering

Antibodies conjugated to bioactive compounds allow targeted delivery of therapeutics to cell types of choice based on that antibody’s specificity. Here we develop a new type of conjugate that consists of a nanobody and a peptidic ligand for a G protein-coupled receptor (GPCR), fused via their C-termini. We address activation of parathyroid hormone receptor-1 (PTHR1) and improve the signaling activity and specificity of otherwise poorly active N-terminal peptide fragments of PTH by conjugating them to nanobodies (VHHs) that recognize PTHR1. These C-to-C conjugates show biological activity superior to that of the parent fragment peptide in vitro. In an exploratory experiment in mice, a VHH-PTH peptide conjugate showed biological activity, whereas the corresponding free peptide did not. The lead conjugate also possesses selectivity for PTHR1 superior to that of PTH(1-34). This design approach, dubbed “conjugation of ligands and antibodies for membrane proteins” (CLAMP), can yield ligands with high potency and specificity.

A ntibodies bind tightly and specifically to their targets, even in highly complex environments. This property of antibodies has been used to deliver bioactive compounds to sites of interest, both for diagnostic and therapeutic applications 1 . For example, conjugates between antibodies and cytotoxic drugs (antibody-drug conjugates or ADCs) can selectively kill cancer cells that display the antibody's target 2 . The success of ADCs often depends on the internalization of the conjugate through endocytosis, followed by release of the cytotoxic payload. Far fewer studies have made use of antibodies to deliver bioactive compounds with sites of action at the cell surface. The conjugation of a ligand for a surface receptor to an antibody that recognizes that same receptor should increase the effective concentration of the ligand and so increase its potency and specificity, provided appropriate spatial constraints are maintained. Ideally, this method could be used with an antibody that directly targets the receptor of interest to enable application without the need for genetic modification of the target cells or organism. The G protein-coupled receptor (GPCR) family of proteins is an attractive class of targets to pursue using this approach.
Molecules that target GPCRs represent more than 25% of all approved drugs 3 . Antibodies and the variable fragments of camelid heavy chain-only antibodies (VHHs or nanobodies) have found increasing use for modulating GPCR signaling 4,5 . GPCRs and their ligands display a considerable degree of degeneracy. Several natural ligands bind to more than a single GPCR and many GPCRs can bind more than one ligand 6,7 . The parathyroid hormone receptors constitute one such example: a bioactive N-terminal fragment of parathyroid hormone (PTH, residues 1-34), used under the name teriparatide to treat osteoporosis, potently activates both type-1 and type-2 PTH-receptors (PTHR1/PTHR2) 8 . PTHR1/2 are part of the class B of GPCRs which are naturally activated by large (>25 residue) peptides 9 . Despite intense industry interest, no small molecule agonists of B-family GPCRs with potencies comparable to the natural ligands have been described. To address PTHR signaling and selectivity, we prepared conjugates of fragments of PTH and VHHs. VHHs are appealing building blocks for these conjugates, as they are the smallest antibody fragments that retain the ability to bind antigens. They can be produced in high yield recombinantly 10 . The site of antigen recognition on VHHs is near the N-terminus 11,12 and the interaction of PTHR1 and PTHR2 with their ligands requires a free N-terminus on the latter 8,13,14 . Using a chemo-enzymatic approach we therefore made C-to-Cterminal fusions of PTH fragments and VHHs 15 to avoid any Nterminal obstructions. These chimeric molecules, dubbed: conjugates of ligands and antibodies for membrane proteins or CLAMPs, target either wild-type or engineered receptor variants (Fig. 1). The optimized CLAMPs display biological activities in vitro and in vivo that are superior to those of the PTH fragments from which they were derived. When otherwise weakly active PTH fragments are incorporated into these conjugates, they can be made exquisitely selective for activation of only those receptors engaged by the VHH. This stands in marked contrast to the lack of selectivity shown by PTH   8 . These findings  suggest that CLAMPs should be broadly applicable for the design of ligands with unique and useful properties.

Results
Receptor constructs and conjugates used for targeting. PTH  interacts with PTHR1 via a two-site mechanism of interaction ( Fig. 1a, b) 8,16 . The association between the extracellular domain of PTHR1 and residues 15-34 of PTH provides the bulk of the binding energy and specificity for this interaction. The association between the transmembrane domain of PTHR1 and residues 1-14 of PTH induces a conformational change in the receptor, which initiates intracellular signaling cascades. This mode of interaction, supported by a large amount of structure-activity relationship data, has been confirmed recently by crystallography and cryo-electron microscopy of PTHR1-ligand complexes (Fig. 1a) 13,14 .
To mimic receptor association exhibited by PTH(1-34), we used either wild-type PTHR1 or PTHR1 variants modified to carry an epitope in the extracellular domain recognized by a VHH of choice. While there is no structural information for any VHH bound to PTHR1, we envisioned a mode of interaction between the receptor and VHH-PTH conjugates like that depicted in Fig. 1c. The portion of PTHR1 encoded by exon 2 is not resolved in structural studies ( Fig. 1d) 13,14,17 , is not important for ligand binding 18 , and in past work has been targeted as a site for receptor modification 18,19 . We generated a construct that encodes a PTHR1 variant in which a 14-residue fragment from exon 2 was replaced with a 14-residue epitope tag from the intracellular protein UBC6e (PTHR1 6E , Fig. 1d) 20 . We also used a receptor construct in which a pH-sensitive green fluorescent protein variant (GFP) was inserted into the portion of the receptor encoded by exon 2 (PTHR1 GFP in Supplementary  Fig. 1) 19 . Another version of PTHR1 in which yellow fluorescent protein (YFP) replaces the entire N-terminal extracellular domain (PTHR1 YFPΔECD , Fig. 1e) 21 , was also used.
To target these receptors, we constructed conjugates comprised of N-terminal fragments of PTH and VHHs (Fig. 1). We used VHHs that recognize green or yellow fluorescent proteins (VHH GFP ) 22 , a 14-mer peptide fragment from the intracellular protein UBC6e (VHH 6E ) 20 , or PTHR1 itself (VHH PTHR ) 23 . We expressed C-terminally His-tagged VHHs in bacteria in a form amenable to subsequent site-specific functionalization at the Cterminus, using sortase A-mediated labeling (sortagging) 24,25 . To these purified VHHs we attached either a triglycine-modified fluorophore for cytofluorimetry or a peptide with azide and biotin moieties for biorthogonal chemistry and conjugate tracking, respectively (Fig. 2).
PTHR1 peptide ligands and conjugation to antibody fragments. To test whether delivery of PTH fragments to their site of action by conjugation to VHHs affects their signaling activity, we synthesized N-terminal fragments of PTH (Fig. 2, Table 1, Supplementary Fig. 6). These fragments were prepared as C-terminal amides by conventional solid-phase peptide synthesis, and purified. Their identities were confirmed by mass spectrometry   Supplementary Fig. 2). Most of these peptides contained several of the modifications found in the M-PTH series of PTH peptides, including the non-standard residue aminoisobutyric acid (Aib) at position 3, which enhances the biological activity of these short PTH fragments (Fig. 2) 29 . Each of these peptides contained a C-terminal cysteine (Cys). Using Cys-maleimide chemistry we appended a dibenzylcyclooctyne (DBCO) handle (Supplementary Fig. 2) to enable an azide-alkyne conjugation between the C-termini of an azide-functionalized VHH and a DBCO-modified synthetic peptide. Of note, the resulting triazole linkage is not susceptible to cleavage by reduction, unlike the disulfide linkages used in other conjugates. The composition of the conjugates was confirmed by mass spectrometry (Fig. 2, Supplementary Fig. 3). For comparison, we also prepared conjugates in which a PTH (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14) analog with an N-terminal triglycine extension (G 3 -PTH(1-14)) was conjugated to VHHs using sortagging, resulting in a conjugate with the more conventional C-N configuration ( Supplementary Fig. 7). We then assessed the capacity of these peptides and conjugates to stimulate the production of cyclic adenosine monophosphate (cAMP), a second messenger molecule produced upon PTHR1 activation, using HEK293 cells expressing a PTHR variant targeted by the relevant VHH and a luciferase-based cAMPresponsive reporter 30 . Progressive truncation of C-terminal residues from PTH(1-34) caused a marked loss in the potency on wild-type PTHR1 and other PTHR1 variants with intact ECDs (Table 1, Supplementary Fig. 6). Addition of a triglycine appendage at the N-terminus of PTH(1-14) caused a reduction in potency, relative to PTH(1-14) with a free N-terminal amine ( Supplementary Fig. 7), in line with precedent 31 . Conjugates in which G 3 -PTH(1-14) was ligated to the VHH C-terminus using sortase were completely inactive, emphasizing the importance of a free N-terminus for PTH and its fragments ( Supplementary  Fig. 7). In contrast, conjugates formed by C-to-C-terminal fusion were active (Table 1, Supplementary Fig. 6).

Discussion
Antibodies as part of conventional antibody-drug conjugates deliver cytotoxic compounds that typically target intracellular proteins 1,2 . Less explored is the use of antibodies to deliver ligands for surface receptors such as GPCRs. This is likely due to complications in preparing homogenous and bioactive conjugates of antibodies and ligands that rely on antibody binding to potentiate engagement of receptor by the ligand. Immunocytokines (conjugates of cytokines and antibodies) are an exception 35 . Immunocytokines have progressed to the clinic, but not without toxicity, suggesting an inadequacy in targeting 35 . In one case, even the identity of the targeting antibody of the immunocytokine was irrelevant for its in vivo efficacy 36 . Mutations in cytokines, introduced to improve the selectivity of immunocytokines, can dampen the affinity for their receptors 37,38 . Immunocytokine-based approaches differ from the CLAMP platform in that they rely on genetic fusions (and genetically encoded residues) and the use of full-size cytokine domains, as opposed to the small peptide fragments with non-natural residues used here.
Agonist activity for VHH-PTH conjugates was completely dependent on binding of the VHH to the receptor being targeted: a mismatch between specificity of the VHH and the receptor construct led to a loss in conjugate activity. We identified a conjugate, VHH PTHR -PTH (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14), with very potent signaling activity in cell-based assays (Table 1), with biological activity in mice (Fig. 5), and with selectivity for PTHR1 over PTHR2 that far surpasses the selectivity of PTH(1-34), the prototypical PTHR1 agonist (Fig. 4) used clinically. PTHR1 mediates the biological activity of PTH in treating osteoporosis, whereas the function of PTHR2 is more obscure. Tools to selectively target PTHR1, and subtypes of GPCRs in other families, will be useful for dissecting the biological function of receptors for which potent and selective ligands are scarce. Success in targeting PTHR1 over PTHR2 sets the stage for designing ligands that specifically activate other receptors with overlapping specificities 6,7 .
The ability to deliver ligands to specific subtypes of receptors, or to receptors engineered to contain an antibody-recognized tag, should allow the creation of (modular) versions of designed receptors exclusively activated by designer drugs (DREADDs) 43 . Previously described DREADDs for GPCRs were identified through modification of the ligand binding site of naturally occurring GPCRs, so that the modified receptors respond to a designer small molecule but not the ligand of the prototype receptor. These designer molecules selectively activate the designer receptor but not any endogenously expressed alternative 44,45 . A similar approach has been deployed to produce an orthogonal receptor-ligand pair for interleukin-2 46 . Our finding that VHH GFP -PTH(1-11) potently activates PTHR1 YF-PΔECD (EC50~0 .15 nM) but is inactive at wild-type PTHR1, suggests a path toward using VHH-tag recognition as a way to convert a GPCR of choice into a DREADD. One aspect of GPCR pharmacology that has not been faithfully reproduced in some DREADD constructs is that of ligand binding kinetics 47 . For some receptors, such as PTHR1, the duration of ligand binding and the signaling induced as a consequence can dictate the type of physiological response evoked. The duration of the cAMP response elicited by PTHR1 activation is correlated with the strength and duration of the calcemic response in vivo 29,48 . Several of the VHH-PTH conjugates tested here induce cAMP signaling that is prolonged relative to the free peptide and similar to that of PTH(1-34) (Supplementary Fig. 8), suggesting that the affinity provided by VHH binding can be used as an independent means to adjust ligand binding and signaling kinetics.
In conclusion, we show that the conjugation of otherwise suboptimal PTHR1 agonist peptides to VHHs that target the intended receptor provides a substantial increase in agonist potency and receptor selectivity. The ability to modulate receptor affinity while not modifying the structure of the agonist used to activate signaling should enable a further dissection of connections between ligand affinity, receptor signaling kinetics, and ligand bias 49 . Preliminary analyses suggest that VHH-ligand conjugates can be designed that possess signaling properties that diverge from that of the natural ligands (Table 1, Supplementary  Fig. 13). The CLAMP platform should be amenable to targeting other GPCRs, especially those with large peptide ligands that bind to their receptors via a two-site mechanism, such as family B GPCRs and chemokine receptors. Efforts are underway to expand this platform to other GPCR-ligand systems. The applicability of this platform will expand as VHHs that bind to new targets on the cell surface are discovered 5 .

Methods
General. HEK293 cell lines (ATCC) were cultured in DMEM medium supplemented with 10% (v/v) fetal bovine serum and penicillin/streptomycin. Cell lines were routinely tested for mycoplasma infection. LC/MS was performed on a Waters Xevo Q-Tof system equipped with HPLC-C8 columns. Mass spectra were obtained using Q-Tof mass spectrometry with a positive ionization mode. Masses for VHHs and conjugates were calculated via analysis of multiply charged ions using the MaxEnt feature on MassLynx software. Protein and peptide concentrations were calculated using absorption at 280 nm for VHHs and peptides with tryptophan (Trp) residues. For peptides without Trp, the amount of peptide was quantified gravimetrically assuming that the weighed mass consisted of 50% peptide (w/w). Antibody used for staining PTHR1-HA (anti-HA-AF594) was purchased from Biologend (BioLegend #901511) and used at a final concentration of 20 nM. Transfections of HEK293 were performed using Lipofectamine2000 using manufacturer instructions.
Peptide synthesis. Peptides were prepared using conventional solid-phase synthesis methods with Fmoc-protection of backbone amines. Synthesis was performed on Rink-amide linker resin to yield C-terminal amides. Backbone deprotection was performed via treatment with piperidine in dimethylformamide (DMF, 20% vol/vol) for 15 min at room temperature. Coupling was performed using Fmoc-protected amino acids (4 equivalents), N,N,N′,N′-Tetramethyl-O-(1H-benzotriazol-1-yl)uronium hexafluorophosphate (HBTU, 4 equivalents), and diisopropylethylamine (DIPEA, 8 equivalents) in DMF for 45 min at room temperature. Fmoc-Lys(biotin)-OH and Fmoc-Lys(azide)-OH were from used from commercial sources without alteration of the synthetic methods described above. Following completion of synthesis, the resin was dried and deprotection was carried out using a solution of 92.5% trifluoroacetic acid, 5% H 2 O, and 2.5% TIPS. Peptides were precipitated into diethyl ether, pelleted by centrifugation, dried under a stream of air, purified using reversed-phase C18 HPLC using a water-acetonitrile gradient, and lyophilized. The identity and approximate purity of peptides was confirmed by LC/MS (Supplementary Fig. 1). Purified products were dissolved in water (10 mM stock concentration) and stored at −20°C.
Purified peptides with C-terminal cysteines were subjected to a reaction with a 2-fold molar excess of either DBCO-maleimide (Click Chemistry Tools) or DBCO-PEG 3 -maleimide (ConjuProbe) (Supplementary Fig. 10) in solvent with 50% (v/v) dimethylsulfoxide (DMSO) and 50 mM pH 7.4 phosphate buffer and purified by reversed-phase C18 HPLC. The identity of peptides was confirmed by LC/MS ( Supplementary Fig. 2). Purified products were dissolved in DMSO (1 mM stock concentration) and stored at −20°C.
Protein expression and purification. The production and purification of VHH GFP (named VHH-enhancer) and VHH 6E (named VHH05) has been described previously 20,22 . The sequence for VHH PTHR was acquired from the literature (named 22A3) 23 . Although several VHHs that bound PTHR1 were reported, we chose 22A3 as it was reported to have the highest affinity 23 . Briefly, VHHs were expressed using the pHEN6 vector. Plasmids coding for PelB-VHH-LPETGG-His 6 were transformed into WK6 E. coli using heat shock. Transfected WK6 E. coli were grown in Luria Bertani broth under ampicillin selection at 37°C until an optical density at 600 nm between 0.6 and 0.8 was reached. Protein expression was induced by the addition of 1 mM IPTG and cells were grown at 30°C overnight. The bacteria were pelleted by centrifugation and resuspended in TES buffer (50 mM Tris, 650 μM EDTA, 2 M sucrose, 15 mL buffer per liter of culture) to prepare for osmotic shock. After incubating for 2 h at 4°C, 75 ml distilled H 2 O was added, and the bacterial suspension was incubated overnight at 4°C. The bacteria were again pelleted and VHHs were purified from the supernatant by Ni-NTA bead batch purification, followed by buffer exchange. Sortase-A pentamutant was expressed and purified as previously described 3 .
Flow cytometry. Suspensions of cells in PBS were stained for 1 h on ice in the presence of indicated concentrations of VHH probes functionalized with Alexa-fluor647. Cells were pelleted by centrifugation and washed with PBS prior to analysis by flow cytometry (BD Accuri C6). To select intact cells gating was performed on forward scatter/side scatter profiles for analysis (see Supplementary  Fig. 4 for an example of the gating strategy). Data were analyzed using FlowJo version 7.6. The median fluorescent intensity (MFI) of stained cells was used to generate VHH binding dose response curves (Supplementary Fig. 4). For curves that did not reach plateau at the highest concentrations tested, curves were constrained by setting the maximal plateau value equal to that seen when staining that cell line with other VHHs that did achieve a plateau.
Microscopy. Monolayers of HEK293 cells grown on glass cover slips at approximately 80% confluency were washed with Hanks balanced salt solution supplemented with 10 mM HEPES pH 7.4 and 0.1% (w/v) bovine serum albumin (HB). The cells were then incubated with peptide, VHH, or VHH-peptide conjugates in HB at 4 or 22°C for 30 m. After staining, cells were washed with HB three times, and fixed with 4% formalin either immediately after rinsing or following a 15 m incubation at 22°C in DMEM + 10% FBS. Indicated slides were permeabolized using 0.5% Triton X100, followed by staining with commercial anti-HA-Alexafluor 594 antibody (20 nM, BioLegend #901511). Cells were then rinsed and mounted with Vector-shield containing DAPI (to visualize nuclei) on glass slides for imaging. Images were acquired using a Nikon Eclipse Ni system with a ×40 PLAN FLUOR 0.75NA DIC M/N2 objective.
Measurement of cAMP response. These assays were performed as previously described 26 . Briefly, HEK-293-derived cell lines that stably express the Glosensor cAMP reporter (Promega Corp.) 30 and PTHR1, a PTHR1 variant, or PTHR2 were seeded into white sided 96-well plates (50,000 cells/well) and grown to confluency. Confluent monolayers of cells were pre-incubated with CO 2 independent medium containing D-luciferin (0.5 mM) at 37°C until a stable baseline level of luminescence was established (20 min). Varying concentrations of ligands were then added, and the time course of luminescence response was recorded using BioTek plate reader. The maximal luminescence response (observed 12-16 min after ligand addition) was used to construct dose-response data sets (Table 1, Supplementary  Fig. 6).
For the measurement of cAMP signaling duration experiments ( Supplementary  Fig. 8) were performed as previously described 27 . Cells were treated with ligands at the indicated concentrations for 12 min (ligand-on phase). After this period, the medium in each well was removed and the cells were rinsed twice with CO 2independent medium to remove unbound ligand. After the addition of D-luciferincontaining fresh medium to each well, the luminescence was recorded for an additional 30-40 min using a PerkinElmer Envision plate reader (ligand-off phase).
Measurement of cytoplasmic calcium mobilization. The mobilization of Ca 2+ levels was assessed in the HEK293 cell line stably transfected with human PTHR1. Intracellular Ca 2+ levels were assessed using a cell-permeant Ca 2+ sensor, Fura2-AM (Invitrogen). Cells in a black 96-well plate were loaded with Fura2-AM in the presence of Pluronic F-127 for 45 min and then rinsed with Hanks buffered saline solution (HBSS). Following an additional 30-min incubation in HBSS, the plate was analyzed using a PerkinElmer Life Sciences Envision plate reader to monitor fluorescence emission at a wavelength of 510 nm, upon excitation at wavelengths of 340 and 380 nm. The data were recorded at 2-s intervals prior to and after ligand addition. The data were calculated as the ratio of the fluorescence signal obtained with excitation at 340 nm to that obtained with excitation at 380 nm.
Measurement of internalization using GFP fluorescence. Receptor internalization was assessed in the HEK293 cell line stably transfected with human PTHR1-pHluorin2-GFP (GPG10) 19 . Confluent monolayers of cells in black walled 96-well plates were incubated in HBSS with bovine serum albumin (0.1% w/v) and HEPES buffer (pH 7.4, 10 mM). Peptides or peptide-VHH conjugates were added and wells were analyzed by recording fluorescence readouts with excitation at 485 or 405 nm and emission at 535 nm. Data were analyzed as a ratio of fluorescence intensity following excitation at 485/405 nm over the course of 90 min.
Animal experiments and measurement of in vivo response. Mice (CD1 female, age 11 weeks) were treated in accordance with the ethical guidelines adopted by Massachusetts General Hospital. Calcemic response assays were conducted using cohort sizes comparable to past work 26 , which provided data adequate for identifying differences in the time course and magnitude of PTH-induced calcemic responses. Peptides and conjugates were administered at doses that allowed for differentiation between compounds with differing levels of in vivo activity 26 . Statistical analyses were performed assuming Gaussian distribution of data. Mice (n = 4 per compound) were injected subcutaneously with vehicle (10 mM citric acid/ 150 mM NaCl/0.05% Tween-80, pH 5.0) or vehicle containing PTH or conjugate at a dose of 35 nmol/kg body weight. Prior to injection, mice were grouped according to basal blood calcium concentrations to ensure each group possessed similar average (mean) blood ionized calcium levels at t = 0. Blood was withdrawn just before injection (t = 0) or at times thereafter. Tail vein blood was collected and immediately analyzed. Blood Ca 2+ concentration was measured with a Siemens RapidLab 348 Ca2+/pH analyzer.
Data calculations. Data were processed using Microsoft Excel and GraphPad Prism 6. Data from cAMP dose-response assays were analyzed using a sigmoidal dose-response model with variable slope. Data sets were statistically compared by using Student's t test (two-tailed) assuming unequal variances for the two sets.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.