Biased Agonism and β-Arrestin Signaling, Explained

Receptor pharmacology

Biased agonism is the claim that two drugs can switch on the same receptor and still produce different biology — that a receptor is not a simple on/off switch but a junction box, and that a well-chosen ligand can send the signal down one wire while leaving another quiet. For about fifteen years this was one of the most exciting ideas in receptor pharmacology, and the peptide field was right at the centre of it. It is also the idea that has taken the hardest beating from replication studies. Understanding both halves of that story — the mechanism and the correction — is the fastest way to become a sceptical reader of any paper that claims a “biased” peptide.

What biased agonism actually claims

A G-protein-coupled receptor does not talk to one partner. It talks to heterotrimeric G proteins (Gs, Gi/o, Gq, G12/13), to GPCR kinases (GRKs) that phosphorylate its intracellular tail, and to the two non-visual arrestins. Those arrestins are worth naming carefully, because the literature uses two systems interchangeably: β-arrestin-1 is arrestin-2, and β-arrestin-2 is arrestin-3 (gene symbol ARRB2). The “β” is purely historical — the protein was named in 1990 for its action on the β₂-adrenergic receptor, which along with rhodopsin was one of only two purified GPCRs available at the time.

The classical picture is a straight line: agonist binds, G protein signals, GRKs phosphorylate the tail, β-arrestin is recruited, signalling is desensitised and the receptor is dragged into a clathrin-coated pit. Biased agonism proposes that this line can be bent. If a ligand stabilises a receptor conformation that couples well to G protein but poorly to β-arrestin, you might in principle keep the therapeutic effect and drop the side effect — provided the side effect really does run through the arrestin arm. That proviso is where the whole field eventually broke.

The measurement problem, which is bigger than most readers realise

Before believing any bias claim, it helps to know exactly what a bias factor is. The standard tool is the operational model of Black and Leff (1983), which describes an agonist with two numbers: an affinity constant K_A and a transducer constant τ that folds together receptor number and coupling efficiency. The ratio τ/K_A is a measure of how well a ligand drives one particular readout. A bias factor is then a ratio of ratios: you take log(τ/K_A) for your test ligand in each pathway, subtract a reference agonist’s value in the same pathway, then subtract across pathways — the ΔΔlog(τ/K_A) that Kenakin and Christopoulos formalised in 2013.

Three consequences follow, and all three are routinely glossed over:

• It is always relative to a reference ligand. Change the reference and the number changes. A bias factor is not a physical constant of the drug.
• System bias. Cell lines used for pharmacology usually over-express the receptor. Spare receptors amplify G-protein readouts enormously while arrestin recruitment assays — typically BRET or enzyme complementation, with no amplification step — stay small. Eli Lilly’s tirzepatide paper says this plainly: heterologous lines express receptors “unphysiologically high,” the resulting amplification “hampers translation of in vitro to in vivo pharmacology,” and they had to build low-density HEK293 lines to get around it.
• Observation bias. Which G-protein readout you choose can flip the answer. Kliewer and colleagues pointed out that fentanyl looks arrestin-biased when arrestin recruitment is compared with GTPγS binding — and shows no bias at all when it is compared with inhibition of cAMP accumulation, another G-protein-mediated readout.

The 2013 framework drew three separate published rebuttals in the same journal within months. The metric was contested from the day it appeared.

The phosphorylation barcode: how one receptor routes to different outcomes

The mechanism underneath biased agonism is real and well characterised, even where the therapeutic claims collapsed. GRKs do not phosphorylate a receptor tail uniformly. In 2011, Nobles and colleagues showed that GRK2 and GRK6 lay down different phosphorylation patterns on the same agonist-occupied β₂-adrenergic receptor, and that those patterns induce different β-arrestin conformations with different downstream jobs — a “barcode” that the cell reads.

The µ-opioid receptor version of this is unusually well mapped. Its C-terminal tail carries 11 serine and threonine sites, organised into two rapidly phosphorylated cassettes: ³⁵⁴TSST³⁵⁷ and ³⁷⁰TREHPSTANT³⁷⁹. Morphine, a low-efficacy driver of phosphorylation, produces selective phosphorylation at S375 — enough to blunt signalling, not enough to support internalisation. High-efficacy agonists such as DAMGO and fentanyl additionally drive hierarchical phosphorylation at T370, T376 and T379, a step that requires GRK2/3 and that does pull the receptor inside. Mutating all four sites in the ³⁷⁰TREHPSTANT³⁷⁹ cluster abolishes β-arrestin recruitment and internalisation outright.

The µ-opioid receptor: the field’s most instructive reversal

If bias were going to save a drug class, opioids were the obvious candidate. The therapeutic effect and the lethal effect run through the same receptor, so separating them pharmacologically would be worth a great deal. The chain of reasoning was: Bohn and colleagues’ 1999 Science paper showed β-arrestin-2 knockout mice had enhanced, prolonged morphine analgesia; the 2000 Nature follow-up showed those mice did not develop antinociceptive tolerance; and in 2005 Raehal, Walker and Bohn reported that the same knockouts had reduced respiratory depression and constipation. Build a ligand that avoids β-arrestin, and you should get analgesia without the breathing risk.

What followed is a textbook case of a field correcting itself. Kliewer and colleagues built knock-in mice whose µ-opioid receptors were progressively unable to be phosphorylated (S375A, then 10 and 11 site-to-alanine mutants). Desensitisation was blocked and tolerance was greatly reduced — the arrestin mechanism worked exactly as advertised for tolerance. But, in their own words, “respiratory depression, constipation, and opioid withdrawal signs are unchanged or exacerbated,” which led them to predict that G-protein-biased agonists would still cause severe adverse effects.

Then came the replication work. A deliberately constructed three-laboratory consortium — Jena, Sydney and Bristol, using three different plethysmography systems, independently bred knockout colonies and blinded experimenters — found that morphine and fentanyl depress breathing in β-arrestin-2 knockouts in a dose-dependent way indistinguishable from wild-type mice. Constipation showed no genotype difference either. An independent group at UCSF reached the same conclusion in eLife the following year, in normoxia and hypercapnia alike.

The proposed explanation for the original discrepancy is mundane and worth remembering as a general lesson: the 2005 knockouts were on a mixed C57BL/6 × 129SvJ background, the backcrossing generations were uncertain, and some experiments used first-generation offspring — and 129SvJ mice appear to show little morphine respiratory depression to begin with.

Gillis and colleagues then supplied the reinterpretation. Measuring oliceridine, PZM21 and SR-17018 carefully, they found all three had low intrinsic efficacy, and that measures of receptor activation, G-protein coupling and β-arrestin recruitment correlated across every ligand tested. If a ligand were truly biased, those measures would come apart. Instead, the improved side-effect profile is explained by partial agonism — an old, unglamorous idea. Note that this remains a live argument: a 2021 Biochemistry paper argues low intrinsic efficacy alone cannot account for everything observed.

What the approved label actually says

The cleanest way to see how the claim fared is to read the regulatory paper trail, all of which is public. Trevena’s own 2018 press release describes oliceridine as “a G-protein biased mu-opioid receptor (MOR) ligand.” FDA’s advisory-committee briefing document repeats that framing and then reports the result: “none of the oliceridine treatment arms demonstrated a significant reduction in the expected cumulative duration of respiratory safety events compared to morphine” — while morphine had produced a greater reduction in pain intensity than all three oliceridine doses in the bunionectomy study. The committee narrowly voted against, a complete response letter followed on 2 November 2018, and approval came after resubmission in August 2020.

Section 12.1 of the approved OLINVYK label — the Mechanism of Action section — reads in full: “Oliceridine is a full opioid agonist and is relatively selective for the mu-opioid receptor… Like all full opioid agonists, there is no ceiling effect to analgesia for oliceridine.” The words bias, β-arrestin and functional selectivity do not appear. The label carries the standard opioid boxed warning for life-threatening respiratory depression, caps the cumulative daily dose at 27 mg because of QTc prolongation, and states that use beyond 48 hours has not been studied in controlled trials.

TRV027: a clean test in heart failure, and a clean negative

The peptide field ran its own well-designed trial of the concept. TRV027 is a “biased” ligand of the angiotensin II type 1 receptor, designed to block the harmful vasoconstrictor arm while preserving the β-arrestin-mediated pro-contractility arm — an attractive proposition in acute heart failure. BLAST-AHF was a phase 2b, randomised, double-blind, placebo-controlled dose-ranging study across 72 sites in 12 countries, with 621 patients randomised and 618 treated, infusing placebo or 1, 5 or 25 mg/h for 48–96 hours against a five-component composite endpoint.

The result, verbatim: “TRV027 did not confer any benefit over placebo at any dose with regards to the primary composite endpoint or any of the individual components. There were no significant safety issues with TRV027.” The trial was well run, adequately sized for a phase 2b, and answered the question. Carvedilol is sometimes cited as a marketed β-arrestin-biased β-blocker, but that claim was never proven clinically, and a 2017 Nature Communications paper reporting that Gαi is required for its β₁-receptor “arrestin-biased” signalling undercuts the original G-protein-independent framing.

Where selectivity between receptor states still looks real

None of the above means every claim of ligand-dependent signalling is wrong. Two peptide examples remain well quantified.

The PTH1 receptor. Abaloparatide and teriparatide differ not in G-protein-versus-arrestin bias but in which receptor conformation they prefer — the G-protein-coupled state (R^G) or the uncoupled state (R⁰). In competition binding, the two affinities differ by 1,600-fold for abaloparatide (0.20 nM at R^G vs 316 nM at R⁰), 110-fold for PTHrP(1–36), and only 12-fold for PTH(1–34). Because binding to R⁰ is what produces long-lived signalling, abaloparatide’s strong R^G preference predicts a more transient cAMP response — and it is roughly five-fold more potent on cAMP than PTH(1–34) (EC₅₀ ~0.087 vs ~0.44 nM). A 2021 JBMR Plus paper reports comparable initial pathway engagement across these ligands, so treat this as characterised but contested.

Trafficking bias at the GLP-1 receptor. Jones and colleagues built agonists that deliberately keep GLP-1R at the plasma membrane rather than letting it internalise. Exendin-phe1 — exendin-4 with a single N-terminal histidine-to-phenylalanine swap — recruits β-arrestin poorly and endocytoses less, while remaining a full cAMP agonist. Compared with approved GLP-1 mimetics it produced greater long-term insulin release and better glycaemic outcomes in mice without the corresponding increase in nausea-like behaviour. Here the trafficking consequence, not the label “bias,” is doing the useful work.

Why any of this matters for tolerance, dosing and half-life

Strip away the marketing and biased agonism leaves behind a genuinely useful frame: where a receptor goes after it is activated changes what the drug does over time.

• Desensitisation and tolerance are not the same clock. Receptor desensitisation and internalisation happen in seconds to minutes; analgesic tolerance develops over days to weeks and involves separate adaptations such as adenylyl cyclase upregulation. Blocking arrestin recruitment reduces tolerance without touching those slower adaptations — which is exactly what the knock-in mice showed.
• Tolerance and dependence are separable. In the 2000 knockout work, tolerance was absent while cyclase upregulation and physical dependence persisted.
• Two opposite ways to get a long signal. Keeping a receptor on the surface (the exendin-phe1 strategy) and letting it signal from inside an endosome (the PTH receptor strategy, where internalised receptor–G-protein complexes keep producing cAMP until retromer shuts them down) both produce sustained output — by opposite trafficking routes. “Bias” is not a single design goal.
• Efficacy is the confound to check first. Before accepting any bias claim, ask whether the ligand is simply a weaker agonist, and whether the two pathways were measured with comparable amplification.

Frequently asked questions

Is biased agonism a discredited idea?

No. The molecular machinery — differential GRK phosphorylation, distinct arrestin conformations, pathway-selective outcomes — is well documented. What has been discredited is one specific therapeutic application: the claim that avoiding β-arrestin-2 would separate opioid analgesia from respiratory depression. Generalising from that failure to “bias is fake” is as sloppy as the original overclaiming.

What is a bias factor, in one sentence?

A unitless, reference-dependent log number (ΔΔlog τ/K_A) describing how much more a test ligand favours one pathway over another compared with a chosen reference agonist — which means it is a property of the comparison, not of the molecule.

Are semaglutide or tirzepatide “biased” drugs?

Tirzepatide has a documented imbalance: full agonism at GIPR, partial agonism with very low β-arrestin recruitment at GLP-1R. Whether that causes its clinical performance is a hypothesis; the paper reporting it says the data “suggest” a link. For semaglutide, no biased-agonism claim appears on its label or in its development programme, and its signalling bias is essentially uncharacterised. Treat any confident statement either way as unsupported.

Does any of this affect how a peptide is handled in the lab?

Not directly — bias is a property of receptor signalling, not of the vial. It does affect how you read a datasheet: a compound described as “biased” may simply be a partial agonist, and a potency figure means little without knowing the assay, the receptor density and the reference ligand used.

References

1. Black JW, Leff P. Operational models of pharmacological agonism. Proc R Soc Lond B 1983;220:141–162. royalsocietypublishing.org
2. Kenakin T, Christopoulos A. Signalling bias in new drug discovery: detection, quantification and therapeutic impact. Nat Rev Drug Discov 2013;12:205–216. nature.com
3. Kliewer A, et al. Phosphorylation-deficient G-protein-biased µ-opioid receptors improve analgesia and diminish tolerance but worsen opioid side effects. Nat Commun 2019;10:367. nature.com
4. Gillis A, et al. Low intrinsic efficacy for G protein activation can explain the improved side effect profiles of new opioid agonists. Sci Signal 2020;13:eaaz3140. doi.org
5. Kliewer A, et al. Morphine-induced respiratory depression is independent of β-arrestin2 signalling. Br J Pharmacol 2020;177:2923–2931. bpspubs.onlinelibrary.wiley.com
6. Bachmutsky I, et al. β-arrestin 2 germline knockout does not attenuate opioid respiratory depression. eLife 2021;10:e62552. elifesciences.org
7. US FDA. OLINVYK (oliceridine) prescribing information, NDA 210730. accessdata.fda.gov
8. US FDA. Anesthetic and Analgesic Drug Products Advisory Committee briefing document, oliceridine, 11 October 2018. fda.gov
9. Felker GM, et al. Heart failure therapeutics on the basis of a biased ligand of the angiotensin-2 type 1 receptor (BLAST-AHF). Eur Heart J 2017;38:2364–2373. academic.oup.com
10. Hattersley G, et al. Binding selectivity of abaloparatide for PTH-type-1-receptor conformations and effects on downstream signaling. Endocrinology 2016;157:141–149. ncbi.nlm.nih.gov
11. Willard FS, et al. Tirzepatide is an imbalanced and biased dual GIP and GLP-1 receptor agonist. JCI Insight 2020;5:e140532. ncbi.nlm.nih.gov
12. Jones B, et al. Targeting GLP-1 receptor trafficking to improve agonist efficacy. Nat Commun 2018;9:1602. nature.com

Informational only — not medical advice. VialHelp does not sell peptides and does not recommend any product, protocol or dose. Discuss any medical question with a qualified healthcare professional. Intended for readers 21+.