Pharmacological characterisation of erenumab, Aimovig, at two CGRP responsive receptors

Abstract

Background and Purpose:

Calcitonin gene-related peptide (CGRP) is involved in migraine pathophysiology. CGRP can signal through two receptors. The canonical CGRP receptor comprises the calcitonin receptor-like receptor (CLR) and receptor activity-modifying protein 1 (RAMP1); the AMY₁ receptor comprises the calcitonin receptor (CTR) with RAMP1. Drugs that reduce CGRP activity, such as receptor antagonists, are approved for the treatment and prevention of migraine. Despite being designed to target the canonical CGRP receptor, emerging evidence suggests that these antagonists, including erenumab (a monoclonal antibody antagonist) can also antagonise the AMY₁ receptor. However, it is difficult to estimate its selectivity between receptors because direct comparisons between receptors under matched conditions have not been made. We therefore characterised erenumab at both CGRP-responsive receptors with multiple ligands, including αCGRP and βCGRP.

Experimental Approach:

Erenumab antagonism was quantified through IC₅₀ and pK_B experiments, measuring cAMP production. We used SK-N-MC cells which endogenously express the human CGRP receptor, and HEK293S and Cos7 cells transiently transfected to express either human CGRP or AMY₁ receptors.

Key Results:

Erenumab antagonised both the CGRP and AMY₁ receptors with a ~20–120-fold preference for the CGRP receptor, depending on the cells, agonist, analytical approach and/or assay format. Erenumab antagonised both forms of CGRP equally, and appeared to act as a competitive reversible antagonist at both receptors.

Conclusion and Implications:

Despite being designed to target the CGRP receptor, erenumab can antagonise the AMY₁ receptor. Its ability to antagonise CGRP activity at both receptors may be useful in better understanding the clinical profile of erenumab.

Introduction

Calcitonin gene-related peptide (CGRP) is a neuropeptide with many functions, such as being involved in the cardiovascular system and in metabolism (Russo & Hay, 2023). Strong evidence links CGRP to migraine. For instance, CGRP is widely distributed across the peripheral and central nervous systems with notable expression in the trigeminovascular system, a key region in migraine pathophysiology (Eftekhari & Edvinsson, 2011; Rees et al., 2022). Additionally, infusion of CGRP to migraineurs can cause migraine-like attacks (Lassen et al., 2002). Perhaps most telling is the fact that drugs that dampen the signalling of CGRP can be used clinically to manage migraine (Edvinsson et al., 2018). However, despite the evident success of this drug class, these drugs are not effective in all patients and they can be associated with side effects such as constipation, indicating there is room to improve on current drugs (Alex et al., 2020; Scheffler et al., 2021).

CGRP belongs to a family of structurally related peptides that includes calcitonin, adrenomedullin, and amylin (Chang et al., 2004; Hay et al., 2018; McLatchie et al., 1998). CGRP also exists in two forms, αCGRP and βCGRP (alternatively known as CGRP-I and CGRP-II). Receptors for this peptide family are complex, involving two G protein-coupled receptors (GPCR), the calcitonin receptor (CTR) and the calcitonin receptor-like receptor (CLR), as well as accessory proteins; receptor activity-modifying proteins (RAMPs) 1, 2, and 3. These RAMPs influence the pharmacology of CLR and CTR, changing their preference for ligand binding and subsequent activation (Hay et al., 2018; Poyner et al., 2002). CLR does not traffic to the cell surface in the absence of RAMPs, however, CTR can act as a receptor independently of RAMPs (Christopoulos et al., 1999; Hay et al., 2005; Hay et al., 2018). The receptor classified as the CGRP receptor comprises CLR in complex with RAMP1. In addition to activating this receptor, CGRP is also an effective agonist of the amylin subtype 1 (AMY₁) receptor, which comprises the CTR in complex with RAMP1 (Figure 1 A, B, C) (Christopoulos et al., 1999; Hay et al., 2018; Poyner et al., 2002). In addition to being activated by CGRP, the AMY₁ receptor is also potently activated by amylin (Christopoulos et al., 1999; Hay et al., 2018). Most work to date has focussed on the canonical CGRP receptor; however, increasing evidence suggests that the AMY₁ receptor may also hold relevance to migraine. For instance, administration of pramlintide, an amylin mimetic that potently activates the AMY receptors but only weakly activates the CGRP receptor, can induce migraine-like attacks in patients with a diagnosis of migraine without aura but not healthy volunteers (Ghanizada et al., 2021b). Additionally, subunits of the AMY₁ receptor (CTR and RAMP1) are expressed in migraine relevant structures, such as the trigeminal ganglia and the locus coeruleus (Becskei et al., 2004; Hendrikse et al., 2022; Rees et al., 2022).

Figure 1.

(A) Schematic of the CGRP and AMY₁ receptors, and the relative potency of CGRP and amylin at each receptor. (B, C) Cryo-EM models of αCGRP (yellow) bound to CLR:RAMP1 (blue and pink, respectively; PDB code 6E3Y), and amylin (green) bound to CTR:RAMP1 (orange and purple, respectively; PDB code 7TYF). Erenumab binds to the extracellular portion of these receptors, located within the black dashed box. (D) Zoomed view of the erenumab binding site on CLR:RAMP1 (blue:pink). CTR:RAMP1 (orange:purple) was aligned to CLR:RAMP1 using the cealign command in PyMOL (Schrödinger & DeLano, 2020) and superimposed. Bound peptides are not shown to facilitate visualisation of the erenumab binding site. Residues corresponding to the erenumab binding site on CLR:RAMP1 are shown as sticks, corresponding residues on CTR:RAMP1 are also shown. (E) Amino acid alignment of the CLR and CTR ECD, performed in Geneious Prime using the entire sequence of each protein. A dark coloured background with white text indicates an exact match, a light coloured background with black text indicates a similar residue (using a scoring system of Blosum 70 with threshold of 1). The alignment is annotated with grey boxes above the sequences to identify the erenumab binding site (as determined by Garces et al., 2020; PDB code 6UMG). Amino acids are numbered according to their position within the sequence (CLR above and CTR below). (F) Amino acids 50 – 96 of RAMP1. Grey boxes above the sequence indicate the erenumab binding site. Numbers above the text indicate position within the protein.

There are now eight FDA approved drugs targeting the CGRP axis for the prevention or acute treatment of migraine. Of these, three monoclonal antibodies target CGRP, while the remaining five (ubrogepant, rimegepant, atogepant, zavegepant, and erenumab) target CGRP-responsive receptors. The “-gepants” are small molecule antagonists used as acute and preventative therapies. These drugs have high affinity for the canonical CGRP receptor, and by virtue of their ability to interact with RAMP1, they are also able to act as antagonists of the AMY₁ receptor. For example, rimegepant is approximately ~30-fold selective for the CGRP receptor over the AMY₁ receptor through the canonical cAMP pathway when antagonising αCGRP (Pan et al., 2020).

Both CTR and CLR are class B GPCRs, which are activated by peptides through a two-domain model. The receptor extracellular domain (ECD) facilitates high-affinity binding to the peptide C-terminus, while the juxtamembranous portion of the receptor interacts with the peptide N-terminus to stabilise a receptor conformation that promotes signalling (Karageorgos et al., 2018). Erenumab, a monoclonal antibody, was designed to target the ECD of the canonical CGRP receptor. A crystal structure of erenumab bound to the ECD of the CGRP receptor provide some insight into its antagonism (Garces et al., 2020). Erenumab contacts both CLR and RAMP1 and appears to occlude the ECD binding pocket with which the CGRP C-terminus makes high-affinity interactions (Figure 1 B, D). It is therefore proposed that erenumab functions by preventing receptor binding and activation by CGRP through a primarily orthosteric mechanism (Garces et al., 2020). However, erenumab is much larger than CGRP and does not interact with the receptor in an identical manner; thus defining erenumab as a pure orthosteric antagonist is unlikely to be accurate.

Erenumab can antagonise CGRP receptors expressed endogenously in SK-N-MC cells and in cells transfected with CLR:RAMP1. Current erenumab data is largely derived from IC₅₀ assays. In SK-N-MC cells, erenumab antagonised αCGRP using IC₅₀ assays (Bhakta et al., 2021; Garces et al., 2020; Shi et al., 2016). Likewise, erenumab (or a non-humanised erenumab mimetic) antagonised αCGRP in IC₅₀ assays using oocytes or CHO-K1 cells transfected with CLR:RAMP1 (Hage La Cour et al., 2022; Johnson et al., 2022). This format is dependent on receptor expression levels, the concentration of agonist used, and the dissociation rates of both the agonist and the antagonist making it difficult to compare between studies and systems (Hall & Langmead, 2010). Therefore, these results can only estimate antagonist potency. We do not know whether erenumab exerts competitive or non-competitive antagonism at the CGRP receptor. An ex vivo pK_B study reported erenumab to dose-dependently prevent CGRP-induced relaxation of cerebral vessels, however this model was unable to resolve whether high concentrations of erenumab altered the maximal level of relaxation meaning they could not conclude whether the antagonism was truly competitive (Ohlsson et al., 2019). While ex vivo data provide incredibly useful information, tissue preparations potentially involve mixed populations of receptors. Additionally, we do not know whether erenumab differentially antagonises αCGRP and βCGRP at the CGRP receptor.

Cryo-electron microscopy structures of the αCGRP-bound CGRP receptor and amylin-bound AMY₁ receptor indicate that these two receptors have a high degree of structural homology (Figure 1 A, B, C, D). This information, coupled with the high amino acid sequence similarity between CLR and CTR in the ECD (Figure 1 E) and the shared RAMP1 subunit between the two receptors, suggests that erenumab could also bind to the ECD of the AMY₁ receptor. There is some evidence for erenumab binding to the AMY₁ receptor; erenumab antagonised amylin in HEK293 cells transfected with the AMY₁ receptor in pA₂ assays, and antagonised αCGRP in oocytes transfected with the AMY₁ receptor in IC₅₀ assays (Bhakta et al., 2021; Hage La Cour et al., 2022). This contrasts with data showing that erenumab is unable to antagonise calcitonin in MCF-7 cells which express a mixed population of calcitonin and AMY receptors, and data showing that a non-humanised erenumab analogue could not antagonise amylin at the AMY₁ receptor expressed in CHO-K1 cells (Johnson et al., 2022; Shi et al., 2016). Thus, the ability of erenumab to antagonise the AMY₁ receptor is not resolved.

Collectively, current data suggest that erenumab could target both the CGRP and AMY₁ receptors, however there are missing data. Despite being an FDA approved drug, we still do not know whether erenumab exerts antagonism competitively, or noncompetitively, such as through irreversible binding or allostery. We are also lacking a study which directly compares the ability of erenumab to functionally antagonise multiple physiologically relevant ligands at the CGRP and AMY₁ receptors. We therefore set out to develop a robust pharmacological understanding of erenumab to enable comparisons between previous studies, explore the possible agonist dependency of its antagonism, and develop data that allows comparisons between systems. This work also provides insights into receptor targeted monoclonal antibody antagonist therapeutics; a field which is expanding but where knowledge of pharmacological behaviour for these types of antagonist is not expansive. For instance, antibodies typically bind targets with high avidity and long duration and we do not know whether this influences antagonist profiles.

Materials and methods

DNA constructs

The human CTR, CLR, RAMP1, RAMP2, and RAMP3 constructs were encoded in pcDNA3.1. These constructs incorporated N-terminal tags, being haemagglutinin (CTR, CLR), myc (RAMP1), or FLAG (RAMP2); these tags do not affect signalling (Bailey & Hay, 2006; Qi et al., 2008; Qi et al., 2013). pcDNA3.1 was also used as a vector control in this study. We used the human CT_(a) splice variant of CTR, incorporating the Leu447 polymorphism, which is considered to be the major variant of this receptor. Within this manuscript, CTR will refer to the CT_(a) receptor.

Peptides and antagonists

All peptides were human sequences and were purchased from Bachem (Budendorf, Switzerland). Rimegepant was purchased from MedChemExpress (New Jersey, USA; Cat# HY15498). Erenumab (70 mg/mL) was purchased from MedSurge through the Southern District Health board of New Zealand. Erenumab was received in an injector (as Aimovig^®) and aliquoted directly from the injector into protein LoBind tubes (Medi’Ray, Auckland, New Zealand; Cat# EP0030108116) and stored at 4 °C. CGRP, calcitonin, and adrenomedullin were diluted in sterile H₂O to a concentration of 1 mM. Amylin and rimegepant were dissolved in DMSO to a concentration of 1 mM and 10 mM, respectively. All compounds were stored in protein LoBind tubes at −30 ᵒC. Freeze-thaw cycles were minimised, with a maximum of two per peptide aliquot.

Cell culture and transfection

HEK293S, Cos7, and SK-N-MC cells were used in this study. Cell culture, plating, and transfection were performed as previously described (Bailey & Hay, 2006; Jamaluddin et al., 2022). Cos7 cells (American Type Culture Collection, RRID CVCL_0224), and HEK293S cells (originally provided by Professor David Poyner, Aston University, Birmingham, UK) were used as these cells have previously been shown to lack expression of CLR, CTR, and RAMPs in our hands (Bailey & Hay, 2006; Qi et al., 2013); however, during the duration of this study we detected low levels of functional CTR expression in our HEK293S cells, the impact of which we will explain in our results section. SK-N-MC cells (purchased from ATCC, Cat# HTB-10; RRID CVCL_0530) are typically misclassified as a neuroblastoma cell line however these cells are now regarded to have originated from an Askin’s Tumour and endogenously express receptors consistent with a canonical CGRP receptor (Jamaluddin et al., 2022). Cells were free from mycoplasma contamination as detected by a MycoAlert kit (Lonza, Basel, Switzerland; Cat# LT07–118).

All cells were cultured in complete media containing Dulbecco’s modified Eagle media (DMEM; Cat# 11995065, ThermoFisher) supplemented with 8% v/v New Zealand origin heat-inactivated foetal bovine serum (FBS; Cat# 10372019, Gibco, MA, USA). Cells were grown at 37 ᵒC in a humidified incubator with 5% CO₂, 95% air.

HEK293S and Cos7 cells were passaged to a maximum passage number of 30. Cells were seeded at a density of 15,000 – 22,500 cells per well in 96-well CellBind plates (HEK293S cells, Corning; Cat# COR3300) or 96-well SpectraPlates (Cos7 cells, MediRay; Cat# PEL6005658). After 24 hours, cells were transfected using polyethyleneimine as previously described (Garelja et al., 2022a). Constructs were transfected at a ratio of 1:1 (receptor:RAMP, in terms of plasmid DNA quantity). When investigating the CTR in absence of RAMPs, cells were transfected with CTR:pcDNA at a ratio of 1:1. Following transfection, cells were returned to the humidified incubator and grown for a further 48 hours before being used in experiments.

SK-N-MC cells used in antagonist and agonist characterisation experiments were between passages 51 to 63. Cells were seeded at a density of 20,000 cells per well into 96-well SpectraPlates. Cells were grown for 72 hours before being used in subsequent experiments. DMSO alone was tested in SK-N-MC cells in pilot experiments, and it did not elicit cAMP production at the amount used in our experiments (Figure S1).

cAMP assay

Peptide stimulation was performed as previously described (Woolley et al., 2017). Stimulation media comprised DMEM supplemented with 0.1% bovine serum albumin (BSA; Cat# ABRE-100, MP Biomedicals) and 1 mM 3-isobutyl-1-methylaxnthine made up in DMSO (IBMX; Cat# I5879, Sigma. Overall percentage 0.2% DMSO). Prior to stimulation, existing media was removed from cells and replaced with 50 μL stimulation media. Cells were left to incubate for 30 minutes at 37 °C. During this period peptides and antagonists were diluted in stimulation media. Antagonists were added to the wells, followed immediately by the agonists, except in select experiments where antagonist preincubation was performed. In these experiments, antagonist was added to cells for 15, 30, or 60 minutes at 37 °C before addition of agonist. In all cases, agonist stimulation was performed for 15 minutes at 37 °C, then terminated by aspiration of stimulation media and the addition of ice-cold absolute ethanol. Plates were stored at −30 ᵒC for a minimum of five minutes and a maximum of two weeks before further processing.

cAMP quantification using LANCE and CisBio cAMP Gs Dynamic kits

Due to changes in product availability, two cAMP quantification kits were used; LANCE (PerkinElmer; no longer supplied) and CisBio cAMP Gs Dynamic (CisBio kit; Cat# PEL62AM4PEC, PerkinElmer). To confirm that results from both kits were comparable, cell lysates from the same experiments were tested with both kits. There were no differences between the pEC₅₀ and E_max derived from either kit (Figure S2). For both kits, the ethanol was evaporated from the wells, and cells lysed with detection buffer. Detection buffer comprised 0.35% triton X-100 (Cat# 1086031000, Merck), 50 mM HEPES (4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid) (Cat# H4034–500G, Sigma), and 10 mM calcium chloride (Cat# 1023820500, Merck) at pH 7.4. A standard curve was made in parallel in each experiment according to the manufacturer’s instruction. For the LANCE kit, lysate (5 μL) or standard curve was transferred to 384-well Optiplates and incubated with 5 μL anti-cAMP 647 antibody (1:200 in detection buffer) at room temperature for 30 minutes. A detection mix containing eu-Streptavidin and biotinylated-cAMP (1:4500 and 1:1500, respectively, in detection buffer) was created, and 10 μL added to each well (Bower et al., 2018). For the cAMP Gs Dynamic kit, lysate or standard curve (10 μL) was transferred to 384-well Optiplates and incubated with 5 μL of d2-labelled cAMP and 5 μL of anti-cAMP Cryptate (both 1:20 in detection buffer as per kit instructions). For each kit, plates were then left to incubate for four hours at room temperature before detection using a ClarioStar plate reader (BMG LabTech, Germany). For the LANCE kit, excitation was at 340 nm and emission detected at 665 nm. For the cAMP Gs Dynamic kit, excitation was at 340 nm and two emission wavelengths were measured (665 nm and 620 nm) and the ratio of emissions (665/620) used for data processing. Molar quantities of cAMP in each sample were determined using the standard curve included in each experiment.

Design and statistical analysis - overview

All data are presented as the mean ± SEM of n biologically independent experiments. Combined concentration-response curves were generated for presentation by combining the mean of data points from individual experiments. Each biological replicate consisted of cells from distinct passages being plated, receiving separate transfection mixes (Cos7 and HEK293S cells only), and stimulation performed with distinct dilutions of drugs. Each biological replicate is referred to as an independent experiment and counted as one n. Experiments were replicated a minimum of five times, except in cases where no antagonism was detected, and therefore no statistical analysis performed. In some instances, experiments were replicated more than five times. This is because we performed some experiments before testing our HEK293S and Cos7 cells for endogenous receptor expression. Once it was evident that an endogenous calcitonin-responsive receptor was present, we conducted additional experiments to determine the level of influence that this might have on our curve fits. We included all experiments in our final analysis, leading to unequal group sizes in some instances. Blinding was not performed in this study, however, agonists and antagonists were randomly assigned positions on the plate in each experimental replicate to remove potential bias from plate positions. Each biological replicate consisted of three technical replicates, allowing four eight-point curves per plate. When multiple plates were required to due to the number of conditions being tested (such as agonist testing in SK-N-MC cells), conditions were randomly assigned to plates and plate positions. Each pA₂ or pK_B experiment had a control agonist curve (in the absence of antagonist) and either one (pA₂) or three (pK_B) curves in the presence of antagonist. The data and statistical analysis comply with the recommendations on experimental design and analysis in pharmacology (Curtis et al., 2018).

Curve fitting and data analysis

All data analysis, including curve-fitting and statistical testing, was performed in GraphPad PRISM 9.0 (GraphPad Software Inc. San Diego, CA). In all cases t-tests and post-hoc tests were employed only when was variance was homogenous (data were normally distributed); for ANOVA, subsequent post-hoc tests were only performed when the initial ANOVA reached a statistical threshold (p < 0.05). Statistical analysis was only performed where group sizes equalled or exceeded n=5; this number was chosen as the minimum group to reliably detect statistical significance based on journal guidelines. The group size is the number of biological replicates and statistical analysis used these independent values. Statistical significance was accepted when p < 0.05.

All data used to derive curves were first fit with weighted non-linear regression (Figure 2). These were our initial fits before antagonist analysis. For each independent experiment, we compared whether the data set was better fit by curves with Hill slopes constrained to one or with free Hill slopes. F-tests and visual inspection of how well the curve fit to the data points suggested that Cos7 and SK-N-MC cells were better fit by curves in which the Hill slope was constrained to one, while data from HEK293S cells were better fit by curves in which the Hill slope was not constrained. This was confirmed by investigating residuals from the curve fits. It is likely that this difference is due to the presence of endogenous CTR in our HEK293S cells (Figure S3 A and C) which provided a mixed receptor population with which agonist and antagonists could interact. Therefore, all data from Cos7 and SK-N-MC cells was analysed with three-parameter nonlinear regression, while HEK293S data was analysed with four-parameter nonlinear regression (Figure 2).

Figure 2.

Overview of analysis carried out to determine antagonist activity of erenumab at both the CGRP receptor and AMY₁ receptors in transiently transfected HEK293S and Cos7 cells. *If Hill slope did not have a statistically significant difference, share slopes in global Gaddum-Schild analysis; otherwise do not constrain. See methods for further discussion of our analytical approach.

When a curve in an independent experiment did not reach a plateau within the tested concentration range, the curve-fit maximum was constrained to the mean response at the highest concentration of peptide in that condition. From each biological replicate we derived a pEC₅₀, E_max and (for HEK293S cells) a Hill slope value (see Figure 2). Values from independent experiments were pooled for analysis. These values were statistically compared using one-way ANOVA with Tukey’s post-hoc test.

Method for determining IC₅₀ values

In IC₅₀ experiments, increasing concentrations of antagonist were used against a fixed concentration of agonist. These agonist concentrations were 0.1 nM, 1 nM, 3 nM, or 100 nM αCGRP in SK-N-MC cells, or 3 nM αCGRP at the CGRP and AMY₁ receptors in HEK293S cells. For each independent experiment, an f-test was performed to determine whether data were better fit using three-parameter or four-parameter weighted logistic fits. The majority of experiments preferred three-parameter fits and we therefore applied this fit to all data sets. As such, all individual pIC₅₀ values were determined from independent experiments using a three-parameter weighted logistic fit. Mean pIC₅₀ values were compared using Student’s t-tests (HEK293S cells comparing erenumab pIC₅₀ at CGRP and AMY₁ receptors without preincubation) or one-way ANOVA with Tukey’s post-hoc test (SK-N-MC cells with multiple agonist concentrations, and HEK293S when comparing the effect of preincubation within receptors).

We validated the IC₅₀ experimental design using rimegepant in transiently transfected Cos7 cells (Figure S4). When using 3 nM αCGRP as an agonist, the pIC₅₀ value for rimegepant at the CGRP receptor was 8.03 ± 0.16 (n = 6) and was 6.25 ± 0.24 (n = 5) at the AMY₁ receptor which is similar to previous reports (Pan et al., 2020). When using 3 nM amylin as an agonist, the pIC₅₀ value for rimegepant at the AMY₁ receptor was 7.29 ± 0.39 (n = 5). Our experiments demonstrated that rimegepant was able to elicit complete antagonism of αCGRP, but not amylin, presumably because amylin is also able to act at CTR alone. Based on these results the media alone point was not included when fitting amylin pIC₅₀ data and further pIC₅₀ experiments were not performed with this peptide:receptor combination.

Antagonist potency determination and statistical analysis

We also used experiments in which an eight-point agonist curve was generated in the absence and presence of either one or three concentrations of antagonist. For example, single concentrations of antagonist were used for calcitonin at CTR and the AMY₁ receptor, while multiple concentrations were used at the CGRP and AMY₁ receptors. We first investigated whether antagonists altered the pEC₅₀ (Cos7 and HEK293S), E_max (Cos7 and HEK293S) or Hill slope (HEK293S only) of curves (Table S1; Table S2). We considered antagonism to be present only when an f-test indicated a difference in the pEC₅₀ or E_max of the curve.

Derived pK_B/pA₂ values were compared with statistical tests dependent on the number of groups relevant to the hypothesis being tested. When two groups were compared (e.g. the pK_B of erenumab against αCGRP at the CGRP and AMY₁ receptors) we used a Student’s t-test, when three or more groups were compared (e.g. the pK_B of erenumab against αCGRP, βCGRP, and amylin at the AMY₁ receptor) a one-way ANOVA with post-hoc Tukey’s test was used. pK_B/pA₂ values were only compared within individual analysis paradigms – the pK_B of erenumab antagonising αCGRP at the CGRP receptor derived through Schild plots was compared only to other pK_B values derived from Schild plots, not to pK_B values derived through other analysis techniques.

Empirical measure of antagonist potency (pA₂)

This analytical approach was taken when only one concentration of antagonist was used and gives an estimate to the potency of an antagonist. Antagonism was quantified using the global-Gaddum-Schild equation described below, with the Schild Slope constrained to one. This constraint was used because we lacked multiple concentrations from which to derive the slope of antagonism, and thus assumed a competitive interaction with a slope equal to unity (Kenakin, 2014).

pK_B analysis approaches

pK_B is the potency of an antagonist, and is also known as the equilibrium dissociation constant. This value is less system dependent than pIC₅₀ values, and thus is more easily translatable across studies (Wyllie & Chen, 2007). pK_B assays also provide important information about the nature of the antagonist. To quantify antagonism from pK_B experiments we used three methods (Figure 2): Schild analysis, the method of Lew and Angus (Clark plots), and the global Gaddum-Schild curve-fitting approach in PRISM.

Schild analysis

A Schild plot was constructed for each independent biological replicate. For each experimental replicate an f-test was performed to determine whether the slope had a statistically significant difference from unity. A consensus approach to analysis was then taken, where the most common fit across all experimental replicates was applied to each individual data set for the final analysis. A Schild slope equal to unity suggests competitive antagonism, and thus the derived value represents a true measure of the pK_B. In cases where Schild slope has a statistically significant difference from unity, the derived value was reported as a pA₂ value, which is an estimate of the pK_B (Kenakin, 2014).

Equations used:

$$CR=\frac{K\text{'}}{K}$$

$$\mathit{\text{Log}}\left(CR-1\right)=\mathit{\text{Log}}\left[B\right]-\mathit{\text{Log}}{K}_B$$

Where CR is the concentration ratio, K’ is the EC₅₀ of the agonist in the presence of antagonist, K is the EC₅₀ of the agonist in the absence of antagonist, B is the concentration of antagonist, and K_B is the antagonist equilibrium binding constant.

The method of Lew and Angus (Clark Plots)

The Clark plot equations described by Lew and Angus were applied to each biological replicate and Clark plots were constructed to visualise data (Lew & Angus, 1995). The equation was applied by using the pEC₅₀ from the curves in absence or presence of antagonist. When fitting the equation, we performed f-tests to compare the standard equation to the power departure equation. We again applied a consensus approach to the data analysis. Preference for the Clark plot equation indicated competitive reversible antagonism, while preference for the power departure suggested deviation from competitive antagonism. Therefore, when values were derived from standard equation, the value represents the pK_B, while values derived from the power departure are pA₂ and treated as estimates of the pK_B (Kenakin, 2014; Lew & Angus, 1995).

Clark plot equation:

$$pE{C}_{50}=-\mathit{\text{Log}}\left(\left[B\right]+{10}^{-p{K}_B}\right)-\mathit{\text{Log\hspace{0.17em}}}c$$

In the standard Clark plot equation, pEC₅₀ is the negative logarithm of agonist EC₅₀ in the absence or presence of the antagonist, B is the concentration of antagonist, pK_B is the negative logarithm of the antagonist equilibrium binding constant, and c is a curve fitting constant.

Power departure version of the Clark plot equation:

$$pE{C}_{50}=-\mathit{\text{Log}}\left({\left[B\right]}^m+{10}^{-p{K}_B}\right)-\mathit{\text{Log\hspace{0.17em}}}c$$

The power departure Clark plot allows the slope of the fitted line to differ from unity, this slope is reflected by the value m. Otherwise all values are as described in the standard Clark plot equation.

Global Gaddum-Schild

Global Gaddum-Schild analysis is a curve fitting approach which fits a family of concentration-response curves using a competitive interaction model. We applied this global fitting approach to each biological replicate. When applying the model to HEK293S cells, the E_min, E_max, and Schild slope were shared between all data sets; for data sets using αCGRP as an agonist the Hill slopes were additionally shared between data sets because there was no statistically significant difference in the Hill slopes of αCGRP curves in the absence and presence of antagonist. When applying the model to Cos7 cells, the Hill slope was constrained to 1 (based on initial fits) and the E_min and Schild slope was shared between all curves. Due to variation between the E_max of each curve within each Cos7 replicate, the E_max parameter was not constrained for Cos7 analyses. However, when the E_max values in the presence and absence of antagonist were compared by one-way ANOVA with post-hoc Dunnett’s test comparing E_max in the absence of antagonist to E_max in the presence of antagonist, there was no statistically significant difference for this data set. For each independent replicate, we performed an f-test to determine whether the Schild slope had a statistically significant difference from unity. We again applied the consensus fit (i.e. Schild slope constrained) to all independent replicates from a data set. We then derived pA₂/pK_B values from each replicate. Sharing parameters between curves is the basis of global curve-fitting approaches. Sharing parameters improves the resultant curve fits and reduces the error in derived values (Hall & Langmead, 2010; Motulsky & Christopoulos, 2004).

Equation used:

$$\mathit{\text{Response}}=\mathit{\text{Emin}}+\frac{\mathit{\text{Emax}}-\mathit{\text{Emin}}}{1+10^(\mathit{\text{log}}\left({10}^{-pEC50}*\left(\left(1+{\left(\frac{B}{{10}^{-pA2}}\right)}^s\right)-A\right)*n\right)}$$

Where B is the concentration of antagonist, s is the Schild Slope, A is the agonist concentration, and n is the Hill slope.

Nomenclature of Targets and Ligands

Key protein targets and ligands in this article are hyperlinked to corresponding entries in http://www.guidetopharmacology.org, and are permanently archived in the Concise Guide to PHARMACOLOGY 2021/22 (Alexander et al., 2021).

Results

Erenumab antagonises αCGRP in SK-N-MC cells which endogenously express CGRP-responsive receptors

Our initial goal was to investigate erenumab antagonist potency in assays commonly used in the literature to enable direct comparison of results. The majority of literature data is pIC₅₀ values obtained from antagonism of αCGRP-induced cAMP production at the CGRP receptor expressed endogenously by SK-N-MC cells (Bhakta et al., 2021; Garces et al., 2020; Shi et al., 2016). Before conducting similar experiments, it was first necessary to confirm that our SK-N-MC cell batch contained a CGRP-responsive receptor. Hence, we characterised the cells using a suite of agonists. βCGRP was ~4-fold more potent than αCGRP, which itself was more potent than adrenomedullin and amylin, while calcitonin was unable to stimulate any detectable signalling; this profile is consistent with CLR:RAMP1 expression (Bailey & Hay, 2006; Hay et al., 2018; Pin et al., 2007; Wunder et al., 2008) (Figure S5). SK-N-MC cells are reported to lose expression of CGRP-responsive receptors as they are progressively passaged (Choksi et al., 2002), therefore we investigated agonist responsiveness over a range of passages to determine the boundaries under which we could conduct our experiments. The cells remained CGRP-responsive over the time-frame we passaged them (Figure S5 and S6).

We then proceeded to perform IC₅₀ assays using multiple concentrations of αCGRP. We took this approach because each previous study has used a different concentration of αCGRP, ranging from 0.3 nM to 100 nM (Bhakta et al., 2021; Garces et al., 2020; Shi et al., 2016). The concentrations we used ranged from 0.1 nM to 100 nM, representing an approximate EC₁₀ to supramaximal stimulation of receptors in our SK-N-MC cells (Figure 3 A). The pIC₅₀ of erenumab increased as the concentration of αCGRP used in the assay increased (Figure 3 B), with the pIC₅₀ being 8.85 ± 0.04 (using 0.1 nM αCGRP), 8.66 ± 0.10 (using 1 nM αCGRP), 8.45 ± 0.11 (using 3 nM αCGRP), and 7.24 ± 0.04 (using 100 nM αCGRP), all n = 5. These data were statistically compared using one-way ANOVA with Tukey’s post-hoc test to compare each pIC₅₀ to each other pIC₅₀. Erenumab was more potent against 0.1 nM than against 3 nM and 100 nM αCGRP (p < 0.05; n = 5). Likewise, erenumab was more potent when antagonising 1 nM or 3 nM αCGRP than when antagonising 100 nM αCGRP (p < 0.05; n=5). These results are consistent with literature values.

Figure 3.

A, C. Concentration-response curves for αCGRP stimulating cAMP production in (A) SK-N-MC cells or (C) in HEK293S cells transiently transfected with the CGRP or AMY₁ receptor. Dotted vertical lines indicate the concentration of αCGRP used in subsequent IC₅₀ experiments. (B, D) Concentration-response curves showing antagonism of αCGRP stimulated cAMP production in (B) SK-N-MC cells or (D) HEK293S cells transiently transfected with the CGRP or AMY₁ receptor. HEK293S cells were stimulated with 3 nM αCGRP. Data are the mean ± SEM of 5 independent experiments.

Erenumab antagonises CGRP or AMY₁ receptors in HEK293S cells

We next profiled erenumab at defined receptors in transiently transfected HEK293S cells to compare to work performed previously (Bhakta et al., 2021). We first performed agonist concentration-response experiments to confirm the potency of αCGRP at both receptors (Figure 3 C). We proceeded to perform IC₅₀ experiments to allow comparison to previous literature. We used αCGRP at a concentration of 3 nM, which produced equal and robust cAMP production at both receptors (Figure 3 C). Erenumab was an antagonist at both the CGRP and AMY₁ receptors, being more potent at the CGRP receptor with pIC₅₀ values of 7.61 ± 0.09 (CGRP receptor, n=5) and 5.77 ± 0.20 (AMY₁ receptor, n=5, p < 0.05). This represents an approximate 70-fold difference (Figure 3 D).

Previous work measuring erenumab binding to the isolated ECD of the CGRP receptor reports that the complex between erenumab and the isolated CGRP receptor ECD has a long half-life (Garces et al., 2020). We therefore reasoned that preincubating cells with erenumab may cause an increase in antagonism over time. We performed IC₅₀ experiments in which cells were preincubated with erenumab for 15, 30, or 60 minutes before the addition of αCGRP (Figure 4). There was no statistically significant difference in the pIC₅₀ of erenumab within receptors across time-points therefore we proceeded with no preincubation for all further experiments (Figure 4, Table 1).

Figure 4.

(A, B) Concentration-response curves showing erenumab antagonising 3 nM αCGRP at the (A) CGRP or (B) AMY₁ receptors in transiently transfected HEK293S cells with and without preincubation with erenumab. Cells were preincubated with erenumab for the indicated times before being stimulated with 3 nM αCGRP. Points are the mean ± SEM of 5 independent experiments. (C, D) Plots showing the pIC₅₀ of erenumab derived from individual experiments at the (C) CGRP and (D) AMY₁ receptors in HEK293S cells. Black lines join points from each independent replicate. There were no statistically significant differences in the derived pIC₅₀ within an individual receptor as determined by one-way ANOVA.

Table 1.

pIC₅₀ values for erenumab antagonising cAMP production stimulated by 3 nM αCGRP at the CGRP or AMY₁ receptors in transiently transfected HEK293S cells. Cells were preincubated with erenumab for the indicated time before adding αCGRP.

	No preincubation	15 minutes	30 minutes	60 minutes
CGRP receptor	8.01 ± 0.19	8.08 ± 0.19	7.95 ± 0.17	7.63 ± 0.10
AMY1 receptor	5.66 ± 0.08	5.85 ± 0.19	5.80 ± 0.15	5.84 ± 0.12

Data are the mean ± SEM of 5 independent experiments. There were no statistically significant differences as measured by one-way ANOVA.

We next performed experiments to derive the pK_B in HEK293S cells. Erenumab was a potent antagonist of αCGRP at the CGRP receptor (Figure 5 A, Table 2), and a weaker antagonist of αCGRP at the AMY₁ receptor (Figure 5 B, Table 2). Erenumab was also a weak antagonist of amylin at the AMY₁ receptor (Figure 5 C, Table 2). Erenumab did not alter the Hill slope of αCGRP (Figure 5 D, E), however there was a statistically significant difference between the Hill slope of amylin in the presence and absence of erenumab (Figure 5 F). We hypothesised that this difference could be due to amylin acting on free CTR, as CTR is able to traffic to the cell surface by itself and respond to amylin. This was tested using immunocytochemistry detecting tagged receptor components where we found mixed receptor populations at the cell surface (Figure S7). This finding should be interpreted in conjunction with the endogenous CTR we detected in our HEK293S cells (Figure S3 C).

Figure 5.

Erenumab antagonism at the CGRP and AMY₁ receptors in HEK293S cells. (A, B) are concentration-response curves for erenumab antagonising αCGRP at the CGRP (A) or AMY₁ (B) receptors, (C) shows concentration-response curves for erenumab antagonising amylin at the AMY₁ receptor. Data are the mean ± SEM of 5 (αCGRP experiments) or 8 (amylin experiments) independent experiments. Curves were fit as described in methods. (D, E, and F) are Hill slopes derived from independent experiments. Data were analysed using one-way ANOVA with post-hoc Dunnett’s test, comparing the Hill slope in the presence of each concentration of antagonist to the Hill slope of agonist alone; * indicates a statistically significant difference. (G, H, and I) show the combined Schild plot. Each point represents a concentration ratio from an independent experiment. (J, K, and L) are Clark plots. Each point represents the pEC₅₀ from an independent experiment. For (G-L), the solid black line indicates the line of best fit while the dotted lines and grey shading indicate the 95% confidence intervals of the line of best fit. (M, N, and O) are the same data as presented in A, B, and C but the data have been refit with the global Gaddum-Schild equation.

Table 2.

Derived pA₂/pK_B values of erenumab antagonising αCGRP and amylin in HEK293S cells transfected with the CGRP or AMY₁ receptors.

HEK293S cells	CGRP receptor	AMY1 receptor
Agonist	αCGRP	αCGRP	Amylin
n	5	5	8 or 6 (Schild)
Schild analysis	8.64 ± 0.15	6.58 ± 0.24*	5.84 ± 0.13^
Lew and Angus	8.76 ± 0.13	6.64 ± 0.22*	5.83 ± 0.11^
Global Gaddum-Schild	8.28 ± 0.10	7.04 ± 0.23*	5.74 ± 0.10^

Data are the mean ± SEM of n independent experiments. There is a difference between the n numbers in the amylin:AMY₁ receptor column due to there being two experimental replicates in which the potency of amylin was not decreased in the presence of the lowest concentration of erenumab (1 μM) meaning it was not possible to derive a log[CR-1] value for the 1 μM concentration in those experiments, leaving only two points with which to fit a line of best fit in the Schild plot. We therefore excluded those replicates from our Schild plot analyses. Other analytical approaches were not restricted by these issues, hence the discrepancy in n numbers. Data are pK_B in all cases except for αCGRP at the CGRP receptor analysed through the global Gaddum-Schild equation which is pA₂ (apparent pK_B); the Schild slope had a statistically significant difference from one (being 1.53 ± 0.14, n = 5). The potency of erenumab against αCGRP was compared between the CGRP and AMY₁ receptors by Student’s t-test; statistical significance from αCGRP at the CGRP receptor indicated by * (asterisk). Likewise, the potency of erenumab against αCGRP and amylin at the AMY₁ receptor was compared by Student’s t-test; statistical significance from αCGRP at the AMY₁ receptor indicated by ^ (caret). The unequal number of n between experiments using αCGRP, and experiments using amylin was due to performing some experiments before testing our HEK293S cells for endogenous receptor expression and so additional experiments were conducted to determine the level of influence that this might have on our curve fits.

We then applied three distinct antagonist analyses to our data sets: Schild analysis, the method of Lew and Angus, and the global Gaddum-Schild equation (Figure 5 G–O). Three antagonist analysis approaches were used as it was not known which analysis would best characterise a monoclonal antibody antagonist. Erenumab was more effective at antagonising αCGRP at the CGRP receptor relative to the AMY₁ receptor through all analysis methods (~20–120-fold, depending on analysis method; Table 2). Additionally, erenumab was a more effective antagonist of αCGRP than amylin at the AMY₁ receptor in HEK293S cells through all analysis methods (~5–20-fold depending on analysis method; Table 2). This range in fold-difference arises due to there being three antagonist analyses which gave subtly different results. Through each analysis technique we also investigated whether the antagonism could be defined by a straight line with a slope equal to unity (in other words, competitive antagonism). This was the preferred fit in all cases, except for αCGRP at the CGRP receptor when analysed by the global Gaddum-Schild equation, where the mean slope was 1.53 ± 0.14 (n =5).

To confirm that the observed antagonism was due to binding to receptors and not due to toxicity associated with high antibody concentrations, we performed experiments using calcitonin at the calcitonin and AMY₁ receptors. Erenumab was a weak antagonist of calcitonin at the AMY₁ receptor (pA₂ of 5.34 ± 0.07, n = 5) but was not able to antagonise calcitonin at CTR alone (Figure S8). This lack of antagonism at CTR alone suggests the antagonism seen in our experiments at the CGRP and AMY₁ receptors is due to erenumab antagonising the receptor and not due to off-target effects on cells (Figure S8).

Erenumab antagonises CGRP or AMY₁ receptors expressed in Cos7 cells

We then proceeded to perform work in transiently transfected Cos7 cells, to compare results in a second cell background that does not have endogenous receptor expression in our hands (Figure S3B and D), although mixed populations of CTR and AMY₁ receptors are still present like in the Cos7 cells (Figure S7). In these cells we expanded our agonist repertoire to include βCGRP. Erenumab was a potent antagonist of both αCGRP and βCGRP at the CGRP receptor (Figure 6 A, B; Table 3). Erenumab was a weaker antagonist of αCGRP and βCGRP at the AMY₁ receptor (Figure 7 A–B, Table 3). Erenumab also antagonised amylin at the AMY₁ receptor (Figure 7 C). We applied our suite of antagonist analyses to these data sets (Figure 6 C–H, Figure 7 D–L; Table 3). There was no difference in the ability of erenumab to antagonise αCGRP and βCGRP at the CGRP receptor (Table 3). Similar results were obtained with the small-molecule antagonist rimegepant which we tested to investigate whether there was a difference between small-molecule and monoclonal antibody therapeutics (Figure S9); rimegepant was chosen for comparison because it has previously exhibited agonist-dependent antagonism (Pan et al., 2020). When antagonising the CGRP ligands, erenumab displayed ~30–120-fold preference for the CGRP receptor over the AMY₁ receptor (Table 3). As with the HEK293S cells, this range in the fold difference arises due to difference in outputs from the three analytical approaches used to quantify antagonism. Unlike the HEK293S cells, there was no statistically significant agonist-dependent antagonism detected at the AMY₁ receptor in Cos7 cells. In most cases, data were consistent with a linear plot with a slope equal to unity. The exceptions were αCGRP at CGRP and AMY₁ receptor (Schild slopes of 1.64 ± 0.27 and 1.64 ± 0.03, respectively; both n=5) and βCGRP at the CGRP receptor (2.08 ± 0.69, n=6), all when analysed using the global Gaddum-Schild equation.

Figure 6.

Erenumab antagonism at the CGRP receptor in Cos7 cells. (A, B) are concentration-response curves for erenumab antagonising (A) αCGRP or (B) βCGRP at the CGRP receptor. Data are the mean ± SEM of 5 independent experiments. (C, D) show the combined Schild plot. Each point represents a concentration ratio from an independent experiment. (E, F) are Clark plots. Each point represents the pEC₅₀ from an independent experiment. For (C-F), the solid black line indicates the line of best fit while the dotted lines and grey background indicate the 95% confidence intervals of the line of best fit. (G, H) are the same data as presented in A and B; however, data have been refit with the global Gaddum-Schild equation.

Table 3.

Derived pA₂/pK_B values of erenumab antagonising αCGRP, βCGRP, and amylin in Cos7 cells transfected with the CGRP or AMY₁ receptors.

Cos7 cells	CGRP receptor		AMY1 receptor
Agonist	αCGRP	βCGRP	αCGRP	βCGRP	Amylin
n	5	5	5	6	6
Schild analysis	9.04 ± 0.19	8.65 ± 0.13	7.04 ± 0.22*	6.61 ± 0.13^	6.64 ±0.13
Lew and Angus	9.11 ± 0.17	8.71 ± 0.14	7.06 ± 0.22*	6.63 ± 0.12^	6.62 ± 0.14
Global Gaddum-Schild	8.63 ± 0.37	8.22 ± 0.34	6.68 ± 0.27*	6.74 ± 0.28^	6.59 ± 0.15

Data are the mean ± SEM of n independent experiments. Data are pK_B in all cases except for αCGRP at the CGRP and AMY₁ receptors and βCGRP at the CGRP receptor analysed through global Gaddum-Schild analysis which is pA₂ (apparent pK_B); the Schild slope had a statistically significant difference from one through this analytical method (slopes of 1.64 ± 0.27 [n=5] and 1.64 ± 0.03 [n=5] for αCGRP at CGRP and AMY₁ receptor, respectively, and 2.08 ± 0.69 [n=6] for βCGRP at the CGRP receptor) meaning that the reported value is an estimate of the pK_B. The potency of erenumab against αCGRP was compared between the CGRP and AMY₁ receptors by Student’s t-test. The same analysis was used when comparing βCGRP at the CGRP and AMY₁ receptors, and when comparing αCGRP and βCGRP at the CGRP receptor. When comparing data from the AMY₁ receptor, one-way ANOVA was used; this did not reach statistical significance and thus no post-hoc test was performed. Statistical significance from αCGRP at the CGRP receptor indicated by * (asterisk), statistical significance from βCGRP at the CGRP receptor indicated by ^ (caret). The unequal number of n was due to performing some experiments before testing our Cos7 cells for endogenous receptor expression and so additional experiments were conducted to determine the level of influence that this might have on our curve fits.

Figure 7.

Erenumab antagonism at the AMY₁ receptor in Cos7 cells. (A, B, and C) are concentration-response curves for erenumab antagonising (A) αCGRP, (B) βCGRP, or (C) amylin at the AMY₁ receptor. Data are the mean ± SEM of 5 (αCGRP) or 6 (βCGRP and amylin) independent experiments. (D, E, and F) are the combined Schild plots. Each point represents concentration ratio from an independent experiment. (G, H, and I) are Clark plots. Each point represents the pEC₅₀ from an independent experiment. For (D-I) the solid black line indicates the line of best fit while the dotted lines and grey background indicate the 95% confidence intervals of the line of best fit. (J, K, and L) are the same data as presented in A, B, and C; however data have been refit with the global Gaddum-Schild equation.

Erenumab was a weak antagonist of receptors comprising CLR/CTR and RAMP3, but not CLR/CTR and RAMP2

We then explored whether erenumab had antagonistic activity at complexes comprising CLR or CTR with the other RAMPs (RAMP2, RAMP3). Erenumab was unable to antagonise amylin at the AMY₂ receptor (CTR:RAMP2), or adrenomedullin at the adrenomedullin subtype 1 (AM₁) receptor (CLR:RAMP2). However, erenumab was a weak antagonist of amylin at the AMY₃ receptor (CTR:RAMP3), and of adrenomedullin at the AM₂ receptor (CLR:RAMP3), with pA₂ values of 5.68 ± 0.14 and 6.01 ± 0.21, respectively (both n=5; Figure S10).

Discussion

We report a framework for interpreting the function of erenumab at CGRP-responsive receptors using different assay formats and analysis approaches. Our data show that erenumab can functionally antagonise the CGRP receptor and the AMY₁ receptor with selectivity estimates between receptors ranging from ~20–120-fold.

Previous erenumab characterisation predominantly reports IC₅₀ values. These range widely in SK-N-MC cells, with values from 7.13 to 8.85. This discrepancy likely arises from different concentrations of αCGRP used to stimulate cAMP production as we show in Figure 3 B. Garces et al., (2020) report a pIC₅₀ of 8.73 with 0.3 nM αCGRP as an agonist, this is very similar to our derived pIC₅₀ using 0.1 nM αCGRP (8.85 ± 0.04). Likewise, Shi et al., (2016) report a pIC₅₀ of 8.64 with 1 nM αCGRP, this is essentially identical to our results using 1 nM αCGRP (8.66 ± 0.10). Bhakta et al., (2021) report a higher pIC₅₀ (7.13) using 100 nM αCGRP as agonist; when we used the same concentration our derived pIC₅₀ was 7.24 ± 0.04. This reinforces that agonist concentration can influence conclusions drawn from IC₅₀ experiments (Frizelle et al., 2006; Neubig et al., 2003; Wyllie & Chen, 2007). We also found that erenumab antagonised the CGRP receptor in IC₅₀ experiments in transiently transfected cells, which is consistent with previous work (Hage La Cour et al., 2022).

We tested erenumab at the CGRP receptor in Cos7 and HEK293S cells in experiments designed to determine its pK_B. Our data predominantly suggest that erenumab acts as a competitive antagonist at this receptor. This conclusion is based upon 1) parallel-rightward shifts without suppression of the E_max with increasing concentrations of erenumab, 2) the majority of antagonist analyses preferring a slope which did not have a statistically significant difference from unity. It is however important to note that global Gaddum-Schild analyses determined that data were better fit by a Schild slope greater than one for both αCGRP (HEK293S/Cos7 cells) and βCGRP (Cos7 cells); this aspect will be covered later in the discussion. Competitive antagonism with erenumab is consistent with data from cranial arteries, which showed parallel rightward shifts of αCGRP-induced relaxation with increasing concentrations of erenumab. This model, under the conditions used, was not able to resolve any effect of erenumab on the αCGRP-induced E_max because αCGRP concentrations were not high enough to reach maximal relaxation in the presence of high concentrations of erenumab (Ohlsson et al., 2019). Our results are indicative of a mechanism of action in which erenumab prevents CGRP from interacting with receptors by reversibly binding to and occluding the peptide binding interface (Garces et al., 2020). This is similar to other GPCR-targeting antibodies, such as Gipg013 which competitively antagonises the glucose-dependent insulinotropic polypeptide receptor (Ravn et al., 2013).

Erenumab also antagonised the AMY₁ receptor. In our IC₅₀ experiments using αCGRP, erenumab was 70-fold less potent at antagonising the AMY₁ receptor relative to the CGRP receptor. This is consistent with previous IC₅₀ experiments in transfected cells (Hage La Cour et al., 2022). This finding differs from other IC₅₀ studies, which describe erenumab (or erenumab mimetics) as being unable to antagonise the human AMY₁ receptor, however these studies used either calcitonin or amylin as agonists (Johnson et al., 2022; Shi et al., 2016). Human calcitonin does not bind with high affinity to AMY receptors, and thus is not an appropriate agonist for AMY receptor studies (Christopoulos et al., 1999; Hay et al., 2005; Hay et al., 2018). We did not test erenumab against amylin in IC₅₀ assays because high concentrations of rimegepant could not fully inhibit amylin-stimulated cAMP production (Figure S4). We further consider amylin as an agonist in the context of erenumab later in this discussion. Furthermore, the concentration of erenumab (or mimetic) used may not have been high enough to achieve quantifiable antagonism (Johnson et al., 2022).

In pK_B experiments at the AMY₁ receptor erenumab generally acted as a competitive antagonist, with increasing concentrations of erenumab causing rightwards shifts in the agonist curve without suppression of the E_max. Competitive antagonism was also suggested by Schild and Clark plots preferring slopes equal to unity. The derived pK_B values (being between 6.5 and 7, depending on cell type and analytical method; Table 2 and 3) were consistent with previous pA₂ experiments (Bhakta et al., 2021). In HEK293S cells, erenumab was more potent against αCGRP than amylin; this is consistent with prior data from HEK293S cells showing that αCGRP was more potent than amylin at disrupting erenumab binding to the AMY₁ receptor (Bhakta et al., 2021). However this agonist-dependence was not seen in Cos7 cells (Table 3). There were a few exceptions to the competitive antagonist conclusion. Firstly, global Gaddum-Schild analyses for αCGRP at the AMY₁ receptor in Cos7 cells preferred a Schild slope greater than one. Secondly, erenumab altered the Hill slope of amylin at the AMY₁ receptor in HEK293S cells, meaning that the curves were not parallel, violating an assumption of competitive antagonism.

We suggest that erenumab changing the Hill slope of the amylin curve in HEK293S cells arises from confounding assay system factors, rather than being intrinsic to the properties of antagonism of amylin by erenumab. A Hill slope which differs from one is often caused by a ligand acting on a mixed receptor population. Amylin potently activates the AMY₁ receptor, but is also relatively potent at CTR alone (Bower et al., 2018; Christopoulos et al., 1999). Transiently transfected cell models, such as ours, result in a non-uniform receptor population due to differential uptake/expression of plasmids between individual cells. Some cells express both CTR and RAMP1 (the components of an AMY₁ receptor), while others express only CTR even when RAMP1 has been co-transfected (Figure S7). This is further complicated by CTR trafficking to the cell surface alone, meaning that even cells which express both CTR and RAMP1 are likely to express a mixed population of CTR and AMY₁ on the cell surface (Christopoulos et al., 1999; Hay et al., 2018; Tilakaratne et al., 2000). This results in amylin acting on two distinct receptor sub-populations (CTR alone and the AMY₁ receptor) and thus a Hill slope that differs from one. This is likely to be exaggerated in our HEK293S cells which have endogenous CTR expression (Figure S3) and is not seen with αCGRP because αCGRP is a very weak agonist of CTR alone (Christopoulos et al., 1999; Hay et al., 2005; Hay et al., 2018). The AMY₁ receptor is sensitive to erenumab antagonism, while CTR alone is not (Figure S8). Erenumab therefore causes amylin to act through a more uniform single receptor population, leading to a Hill slope closer to one. The insensitivity of the CTR, AMY₂, and AM₁ receptors to erenumab at our highest tested concentration suggests antagonism reported in this paper was not related to off-target toxicity associated with high concentrations of erenumab, and instead due through erenumab acting on receptors.

We report that erenumab can weakly antagonise the AM₂ and AMY₃ receptors (Figure S10). Previous studies did not detect erenumab binding to these receptors using flow cytometry or antagonism (AM₁ and AM₂ receptors) using IC₅₀ format assays (Bhakta et al., 2021; Shi et al., 2016). This may be due to differences in the sensitivity of the different assays to detect a weak interaction. Our detected antagonism of RAMP3-based, but not RAMP2-based receptors is perhaps not surprising, given that RAMP1 and RAMP3 are closely related, while RAMP2 is less conserved (Barbash et al., 2017; Liang et al., 2020). Similarly, the gepants, and compounds based on these chemical structures, tend to have higher potency at the AM₂ than the AM₁ receptors (Avgoustou et al., 2020; Salvatore et al., 2008; Salvatore et al., 2010). A potential link between the AM and AMY receptors and migraine has been shown. Infusion of either adrenomedullin or the amylin analogue pramlintide into migraine patients are able to produce migraine-like attacks (Ghanizada et al., 2021a; Ghanizada et al., 2021b). However, it is unlikely that erenumab reaches sufficient concentrations in vivo to functionally antagonise AM₂ and AMY₃ receptors, though we lack sufficient data to draw a firm conclusion.

Ideally, we would have compared erenumab to an isotype-matched antibody which did not target CGRP-responsive receptors, however we could not commercially sourced antibodies at concentrations high enough to be viable in our experiments. Given that each individual erenumab antibody has two arms with which to bind receptors, it is possible that some of its antagonism could be due to each individual antibody binding to two distinct receptors and inducing spatial proximity, however we were unable to test this due to commercial research antibodies being of a much lower concentration than erenumab. Further studies investigating this phenomenon, potentially using anti-epitope tag antibodies are warranted.

When comparing different methods of analysing pK_B experiments we found that all analyses generally gave similar results. Comparing between Schild and Clark plot analyses, the Clark plot had the advantage of relying less heavily on the pEC₅₀ of the control curve. This was advantageous in situations where the pEC₅₀ of the agonist in the presence of low concentrations of antagonist was occasionally similar/identical to the pEC₅₀ of the agonist in the absence of antagonist. Therefore, Clark plots should be considered for more widespread use than Schild plots (Hall & Langmead, 2010; Lew & Angus, 1995). The bulk of our evidence suggests that erenumab acts as a competitive antagonist, with a Schild slope of one and with no effect on E_min, E_max, or Hill slope. However, using the global Gaddum-Schild fitting procedure, we noted four conditions in which Schild slopes were statistically greater than one, suggesting non-competitive antagonism. This was in contrast to the other antagonist quantification approaches we used and thus is important to discuss as we expected to see consistency in outcome regardless of analysis method. Specifically, these four conditions were: αCGRP at the CGRP receptor [Cos7 and HEK293S], βCGRP at the CGRP receptor, and αCGRP at the AMY₁ receptor [both Cos7]). This difference was not due to using shared parameters in our global Gaddum-Schild analyses, as refitting select data sets using the global Gaddum-Schild approach with no constraint on E_min/E_max (as used in manual Schild and Clark plots) did not change the conclusions (Figure S11). Future experiments should explore this potential for apparent non-competitive behaviour. A limitation of our study is the concentration range of antagonist used. A greater number of concentrations over a wider range would give more accurate data and could be considered for future studies (Kenakin, 2014). For the remaining discussion we will use average pK_Bs (K_Bs) derived through our three analytical methods. For erenumab against αCGRP at the CGRP receptor this is 8.56 (2.75 nM; HEK293S cells) or 8.93 (1.18 nM, Cos7), while against αCGRP at the AMY₁ receptor this is 6.82 (151 nM; HEK293S cells) or 6.93 (117 nM; Cos7 cells).

Erenumab has an average peak serum concentration in humans of 70–160 nM, with little difference between males and females, or migraine diagnosis (de Hoon et al., 2018; Shen et al., 2022). At face-value, erenumab could be present at sufficient concentrations to at least partially occupy AMY₁ receptors, and thus some of the clinical effects could be attributed to AMY₁ receptor antagonism. However, there are numerous factors that complicate this interpretation, including a lack of understanding of the quantity and distribution of CGRP and AMY₁ receptors at sites relevant to migraine pathophysiology, the concentration and binding kinetics of erenumab at sites of receptor expression, concentrations of CGRP at sites of receptor expression and the concentration of CGRP required to contribute to migraine pathophysiology. Further study is required to determine whether AMY₁ functionally contributes to migraine.

Using the average pK_B of the three analytic methods, erenumab displayed an 80- to 100-fold higher affinity for the CGRP receptor over the AMY₁ receptor (against αCGRP in HEK293S and Cos7 cells, respectively). This is similar to available data for gepants. Using αCGRP as an agonist in pA₂/pK_B experiments, olcegepant is between 130- and 250-fold more potent at the CGRP receptor than the AMY₁ receptor, while telcagepant and rimegepant are 40-fold and 30-fold more potent at the CGRP receptor, respectively (Garelja et al., 2022b; Hay et al., 2006; Pan et al., 2020; Walker et al., 2015; Walker et al., 2018). A crystal structure of erenumab and molecular docking of rimegepant show an overlap of contact residues between the ECD of CLR and RAMP1, indicating they share a similar binding site (Garces et al., 2020; Leung et al., 2021). The similarity in selectivity profiles between a monoclonal antibody and the small molecule antagonists is likely to be reflective of a similar mechanism of action in their binding to receptor ECDs that share RAMP1 (Garces et al., 2020; Leung et al., 2021; ter Haar et al., 2010). These two drug classes are also similar in that, where tested, both appear to behave as competitive antagonists (Pan et al., 2020).

Our data focussed on antagonism of cAMP production, which is the canonical pathway by which CGRP-responsive receptors signal, however it is important to note that antagonism of these receptors can be pathway specific (Walker et al., 2018). Future studies investigating the ability of erenumab and other antagonists (such as the gepants) to antagonise signalling through alternative pathways such as extracellular-regulated kinase (ERK) phosphorylation or β-arrestin may be useful in interpreting the different physiological outcomes associated with these drug classes.

Understanding how erenumab (and the gepants) affect receptor processes such as internalisation and degradation may also hold some answers to physiological questions. Previous studies are conflicting as to whether erenumab is internalised by cells, with one report showing that erenumab is internalised in cells transfected with CGRP-responsive receptors, both in the absence and presence of CGRP, while another reports that erenumab does not internalise in the absence of CGRP, and prevents the internalisation of the CGRP receptor when co-incubated with CGRP (Bhakta et al., 2021; Manoukian et al., 2019). Furthermore, there are multiple ways in which erenumab could be internalised: receptor mediated mechanisms (in that erenumab binds to, and stimulates internalisation of, receptors), internalisation through a “bystander effect” (in that erenumab binds to a CGRP-responsive receptor, and is internalised due to a nearby receptor becoming activated by CGRP which stimulates internalisation of the surrounding membrane), or internalisation due to non-receptor mediated methods (in that erenumab binds to a receptor, which is then internalised through constitutive membrane recycling). Given the high interest in endosomal signalling of CGRP and its relevance to pain transmission, further research is warranted (De Logu et al., 2022; Yarwood et al., 2017). The internalisation of erenumab, and thus its ability to reach endosomal sites, is also of relevance to differentiating between the gepants and erenumab. The gepants are relatively lipophilic, and thus it is possible that they can pass through cell membranes to target endosomal receptors (Mullin et al., 2020). In contrast, any erenumab internalisation is likely to be dependent on interactions with the cell surface membrane, such as by binding to a CGRP-responsive receptor (Bhakta et al., 2021). As such, it is possible that erenumab and gepants could target different pools of CGRP receptors, and we speculate that this could contribute to their additive effects in migraine management (Mullin et al., 2020).

Conclusions

Erenumab pharmacology was generally consistent with that of a competitive reversible antagonist. Erenumab functionally antagonises both the CGRP and AMY₁ receptors, displaying a ~20–120-fold preference for the CGRP receptor over the AMY₁ receptor, depending on the assay format, cells, analysis and agonist used i.e. IC₅₀ or pK_B, and CGRP or amylin. Selecting values from pK_B analysis (more informative than IC₅₀) using αCGRP, this range becomes ~80–100-fold. Our results suggest that AMY₁ receptor antagonism should be considered in interpreting the mechanism of action for erenumab.

What is known

• The monoclonal antibody erenumab was designed to target the canonical CGRP receptor for migraine management.

• It is unclear whether erenumab also targets the AMY₁ receptor, a second CGRP-responsive receptor.

What this study adds

• Erenumab antagonises both CGRP-responsive receptors, displaying ~20–120-fold preference for the canonical CGRP receptor.

• Erenumab appears to behave as a competitive antagonist at both receptors.

Clinical Significance

• The pharmacological profile of erenumab resembles that of small molecule CGRP receptor “gepant” antagonists.

• The AMY₁ receptor could contribute to erenumab mechanism of action.

Abbreviations

AMY₁

amylin receptor subtype 1

CGRP

calcitonin gene-related peptide

CLR

calcitonin receptor-like receptor

CTR

calcitonin receptor

ECD

extracellular domain

GPCR

G protein-coupled receptor

RAMP

receptor activity-modifying protein

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.