The Signal That Wasn't the Whole Story
Promega Corporation
Publication Date: September 2026
Introduction
The cAMP assay came back clean. The dose-response is where it needs to be, and the program is advancing on solid data. The readout answered the question it was designed to answer. What the assay didn’t measure is already shaping where the program goes.
GPCRs are one of the most productive drug target classes in the history of pharmacology. For most of that history, the cAMP readout was standard because the question it answered was the right one: does the compound modulate the receptor as intended, activating or inhibiting its activity? The model was straightforward. A ligand binds, a G protein activates, a second messenger accumulates and a physiological response follows. Beta-arrestin was part of the picture, but as the off-switch: it terminated signaling and drove receptor internalization. That it could also initiate its own independent downstream signaling wasn't yet appreciated. The tools built around that model, (including radioligand binding assays, reporter gene assays and endpoint cAMP readouts) each answered the question researchers were asking at the time.
What the assay didn’t measure is already shaping where the program goes.
What researchers discovered, as programs pushed deeper into GPCR pharmacology, is that the question needed more resolution than the tools could provide. The same receptor, activated by the same ligand, can simultaneously engage a G protein pathway and a beta-arrestin pathway. What wasn't appreciated is that beta-arrestin doesn't just terminate signaling. It initiates its own signaling cascade. A cAMP assay only sees the G protein arm, but it provides no view into what the beta-arrestin arm is doing in parallel.
The Off-Switch That Wasn't
In 2007, James Wisler and colleagues compared sixteen beta blockers at the beta-2 adrenergic receptor and found the answer through a mechanistic difference. Carvedilol blocked G protein-dependent cAMP production while recruiting beta-arrestin and driving its own downstream signaling. At the same receptor and at the same time, it was an inverse agonist on one pathway, an agonist on the other. The cAMP assay had been accurate. It just hadn't been complete. There was a second signal at the same receptor, and the assay readout had no way to detect it.
The cAMP assay had been accurate. It just hadn't been complete.
That paper didn't introduce biased signaling as a concept. What it did was show it operating in a drug that was already on pharmacy shelves, at one of the better characterized receptors in GPCR pharmacology. Carvedilol showed that biased signaling was no longer hypothetical. If two signaling arms could be pharmacologically separated, selectivity had a new dimension.
When the Hypothesis Broke
Carvedilol opened the door. If a single receptor could drive two distinct signaling arms in parallel, characterizing only one wasn't enough. Drug hunters began asking a harder version of the question: not just whether a compound engaged the receptor, but which arm it preferentially engaged, and what that meant.
For some receptors, a clean hypothesis emerged. In opioid pharmacology, G protein signaling was thought to produce analgesia. Beta-arrestin recruitment was linked to tolerance, respiratory depression and constipation. The distinction traced back to beta-arrestin-2 knockout mouse experiments from the early 2000s. Design for G protein bias, the thinking went, and you could separate the benefit from the liability. The hypothesis did not survive contact with the data.
The mechanistic premise fell apart on multiple fronts. Kliewer and colleagues showed in 2020 that the original beta-arrestin-2 knockout mouse experiments that generated the hypothesis couldn't be reproduced. Morphine and fentanyl induced respiratory depression and constipation in knockout mice indistinguishable from wild-type.
Separately, Gillis and colleagues published a systematic evaluation in 2020 of three of the most prominent MOR agonists proposed to be G protein-biased: oliceridine, PZM21 and SR-17018. The finding was definitive. None of them were actually biased. All three had consistently low intrinsic efficacy across both G protein and beta-arrestin pathways. They weren't selectively engaging one arm over the other. All three were partial agonists with a reduced footprint across all signaling, and the improved side effect profiles reflected their reduced activity rather than pathway selectivity.
Cao and colleagues documented a related problem in 2020 from a different angle. They compared nalfurafine, a kappa opioid receptor agonist with a clinically favorable profile, to 42B, a structural analogue differing only in the absence of a hydroxyl group. Both showed G protein bias in the assay conditions used. Both produced antinociceptive and antipruritic effects in mice, and neither caused conditioned place aversion. Unlike nalfurafine, 42B caused motor incoordination and hypolocomotion at the same doses.
Across all three findings, the pattern was the same: the in vitro picture and the in vivo reality diverged.
Across all three findings, the pattern was the same: the in vitro picture and the in vivo reality diverged.
What the Bias Ratio Doesn't Capture
Gillis and Cao are pointing at different problems. Gillis identified a measurement failure: when receptor reserve is high and signaling is amplified, a low-intrinsic-efficacy compound looks fully active in a G protein assay while showing little activity in the unamplified beta-arrestin assay, and the resulting ratio gets misread as bias. The compound isn't selective. The assay conditions created the appearance of selectivity.
Cao identified a translation failure: even when the bias ratio is genuine, it doesn't travel cleanly from a recombinant cell system to a living animal. What the assay reads is shaped by cell type, receptor expression level, G protein stoichiometry and temporal dynamics. The same compound, measured in a different system, at a different time point, against a different reference ligand, can look meaningfully different. The bias ratio isn't fixed. It's contextual.
Kenakin's 2024 review named the frontier directly: bias is easily detected, but translating a simple in vitro bias ratio to the biology of a living system is the harder problem, and it remains unsolved. The in vitro measurement is a starting point for classification, not a prediction of what the compound will do in a living system.
A cAMP readout tells you what G protein signaling looks like under those conditions. What beta-arrestin recruitment looks like, under those conditions or different ones, remains invisible. The signaling arm you didn't measure doesn't disappear. It shows up later, in the outcome the cAMP assay was not designed to see.
Programs that characterize only one pathway are, by definition, working with incomplete information.
The signaling arm you didn't measure doesn't disappear. It shows up later, in the outcome the cAMP assay was not designed to see.
The Signal That Lined Up
What comprehensive characterization looks like in practice became even clearer in 2025. Tirzepatide's clinical performance had raised the question of whether pathway-selective pharmacology was contributing to its superior profile over earlier GLP-1 agents. Rodriguez and colleagues moved beyond the in vitro bias ratio and tested whether the signal held in a living animal.
CT-859, a dual GLP-1R/GIPR agonist engineered to suppress beta-arrestin recruitment while preserving cAMP signaling, outperformed liraglutide at the same molar dose on glucose reduction. In cells, CT-859 kept receptors at the surface where liraglutide drove them inside, measured with the NanoBiT® assay for beta-arrestin recruitment and HiBiT for receptor internalization, both in real time, in living cells. In mice, that signature lined up with a striking in vivo pattern: at 24 hours, liraglutide's effect had faded, and CT-859's hadn't. The authors were careful not to claim more than the data supported. They proposed receptor trafficking as a likely contributor to that durability, and explicitly acknowledged other mechanisms could be involved.
Rodriguez didn't stop at one comparison, or one compound. They tested CT-859 in GLP-1R and GIPR knockout mice to isolate which receptor was driving the glucose-lowering effect at which dose. Central administration characterized its effects on the brain circuits controlling food intake and body weight directly. They compared its weight loss against 60% caloric restriction, and found CT-859 outperformed it even with equal food intake between groups: evidence the drug's effect on body weight wasn't fully explained by appetite suppression alone. When they tested other biased tool compounds, at GLP-1R and independently at GIPR, the same pattern held.
Other groups have reported that biased GLP-1R agonists outperform unbiased ones on glucose regulation. Separate work in beta cell-specific beta-arrestin knockout mice found the same direction of effect through a different route: prolonged GLP-1R agonist exposure increased insulin secretion when beta-arrestin recruitment was removed. CT-859 is one more result pointing the same way, arrived at through knockouts, central dosing and a caloric-restriction control, not a single bias ratio.
Demonstrating that a ligand's bias connects to its efficacy, for any receptor, takes exactly this kind of convergent evidence.
The Question in Its Current Form
Bias is detected easily. Translation is the unsolved problem. Every GPCR program built on a single pathway readout has an open question about the other, not because the other arm always changes the outcome, but because no one can say whether it does or doesn't without looking. The beta-arrestin arm is measurable, and so are the downstream effectors, G protein subtypes and the kinetics of receptor trafficking. The signaling picture is more textured than any single readout can capture.
The tools to begin characterizing both arms exist. GloSensor™ technology measures cAMP continuously, in living cells, rather than at a single endpoint. NanoBiT® beta-arrestin recruitment and HiBiT receptor internalization do the same for the other arm. Watching both continuously, across the same timescale, is the starting point for understanding what a compound is actually doing.
Andrew Zhang, Director of Pharma/Biotech Market Segment at Promega and a former Chemical Biology Director at AstraZeneca, has seen a version of this from inside drug discovery programs. When a downstream readout comes back inconclusive, the program rarely stops. "The problem is not that the program is closed," Zhang says, "but that the decision making becomes paralyzed and the project continues without a clear direction, wasting precious time and resources."
What unblocks that stall isn't a better guess at the downstream readout. It's a clearer view further upstream, closer to the target itself. Beta-arrestin recruitment, G protein activation and receptor internalization each bring the compound's behavior into sharper focus. None of them is the whole pathway, and neither is any pair of them. Together they narrow what a compound is actually doing, and that narrowing is what turns incomplete information into an informed decision.
References
- Wisler, J.W. et al. (2007) A unique mechanism of beta-blocker action: carvedilol stimulates beta-arrestin signaling. Proc. Natl. Acad. Sci. USA 104, 16657–16662. https://doi.org/10.1073/pnas.0707936104
- Raehal, K.M., Walker, J.K.L. and Bohn, L.M. (2005) Morphine side effects in beta-arrestin 2 knockout mice. J. Pharmacol. Exp. Ther. 314, 1195–1201. https://doi.org/10.1124/jpet.105.087254
- Gillis, A. et al. (2020) Low intrinsic efficacy for G protein activation can explain the improved side effect profiles of new opioid agonists. Sci. Signal. 13, eaaz3140. https://doi.org/10.1126/scisignal.aaz3140
- Kliewer, A. et al. (2020) Morphine-induced respiratory depression is independent of beta-arrestin2 signalling. Br. J. Pharmacol. 177, 2923–2931. https://doi.org/10.1111/bph.15004
- Cao, D. et al. (2020) Comparison of pharmacological properties between the kappa opioid receptor agonist nalfurafine and 42B, its 3-dehydroxy analogue: disconnect between in vitro agonist bias and in vivo pharmacological effects. ACS Chem. Neurosci. 11, 3036–3050. https://doi.org/10.1021/acschemneuro.0c00407
- Rodriguez, R. et al. (2025) Biased agonism of GLP-1R and GIPR enhances glucose lowering and weight loss, with dual GLP-1R/GIPR biased agonism yielding greater efficacy. Cell Rep. Med. 6, 102156. https://doi.org/10.1016/j.xcrm.2025.102156
- Willard, F.S. et al. (2020) Tirzepatide is an imbalanced and biased dual GIP and GLP-1 receptor agonist. JCI Insight 5, e140532. https://doi.org/10.1172/jci.insight.140532
- Kenakin, T. (2024) Bias translation: the final frontier? Br. J. Pharmacol. 181, 1345–1360. https://doi.org/10.1111/bph.16335
Learn more about Target-Based Small Molecule Drug Discovery