Comment
https://doi.org/10.1038/s41589-023-01292-8
A community Biased Signaling Atlas
Jimmy Caroli, Alibek Mamyrbekov, Kasper Harpsøe, Sahar Gardizi, Linda Dörries,
Eshan Ghosh, Alexander S. Hauser, Albert J. Kooistra & David E. Gloriam
Biased signaling gives hormones, probes or
drugs distinct functional outcomes via the
same receptor. The Biased Signaling Atlas
(https://BiasedSignalingAtlas.org) provides a
community hub with data and tools to advance
this paradigm, which may yield safer and more
potent drugs.
‘Biased signaling’ is a molecular mechanism that enables specific
ligands to accomplish functional selectivity through a single recep-
tor by steering its signaling into different intracellular pathways
(Fig. 1). This allows physiological processes to be modulated differen-
tially by alternative endogenous ligands. For example, the chemokine1,
opioid2, pituitary adenylate-cyclase-activating polypeptide (PACAP)3,
protease-activated4, serotonin5 and parathyroid hormone (PTH)6 recep-
tor systems have evolved multiple endogenous agonists eliciting biased
signaling. That is, they each have a principal endogenous ligand that
is considered to be balanced, and additional endogenous ligands that
differentially activate receptor coupling at these receptors.
Furthermore, drugs can exhibit biased signaling that lets them
favor signaling pathways associated with therapeutic effects over
those related to adverse effects7,8. Hence, biased signaling opens up
new opportunities to increase drug efficacy and safety and to target
pathways separately. One-third of drugs9 and two-thirds of hormones10
modulate G-protein-coupled receptors (GPCRs). For GPCRs, biased
signaling has classically been studied for the four G protein families (Gs,
Gi/o, Gq/11 and G12/13) and the arrestin family11 (Fig. 1). It additionally could
extend to GPCR kinases (GRKs), which are also transducers — that is,
receptor-binding effector proteins that phosphorylate the C termini
of activated receptors to promote arrestin recruitment11. Recently,
however, the scope of the term has been greatly expanded to include
bias among any of the 27 specific transducer subtypes — 16 G proteins
(with separately signaling α and βγ subunits), 7 GRKs and 4 arrestin pro-
teins — which can differ in their functional outcome12–14. For example, a
new study has demonstrated that the selective adenosine A1 receptor
agonist benzyloxy-cyclopentyladenosine preferentially activates the
Gob G protein subtype and achieves analgesia while avoiding sedation,
bradycardia, hypotension and respiratory depression15.
Biased signaling varies across experimental systems and biosen-
sors, often leading to problems with reproducibility and description.
For example, the Gi-biased activity (over arrestin) of the drug olic-
eridine via the μ-opioid receptor was hypothesized to reduce adverse
effects, but this was later contradicted16. Recent community guide-
lines from the International Union of Basic and Clinical Pharmacology
address this by defining terminology and recommendations for the
design and reporting of ligand bias studies11. Ligand bias is defined as
the ligand-dependent preferential receptor activation of one over other
transducer pathways relative to a reference ligand in a given cellular
nature chemical biology
Check for updates
Biased ligand
(~9,000 data points) Logo
Biased
signaling
atlas
Receptor
Biased signaling
G proteins GPCR kinases Arrestins
n = 16 n=7 n=4
Pathway effects
(~640 functional outcomes or therapeutic areas)
Fig. 1 | Biased signaling. A biased ligand steers receptor signaling towards
one out of several transducer pathways leading to a distinct physiological or
therapeutic effect. Data referred to were curated from 269 scientific articles.
system. Functional selectivity is the observed overall response combin-
ing ligand bias and system bias, which collates all non-ligand-induced
biases inherent to the cellular system—for example, the protein expres-
sion levels. Furthermore, the community guidelines provided clear
definitions for three distinct types of ligand bias depending on the
reference ligand used: (i) ‘pathway bias’ versus a pathway-balanced
reference (that has near-equal pathway signaling), (ii) ‘physiology
bias’ versus an endogenous ligand and (iii) ‘benchmark bias’ versus any
ligand of interest (for example, a drug or often-used tool compound).
Tremendous work lies ahead to identify pathway-selective probes
and drugs that are currently lacking and to map therapeutically relevant
signaling pathways. These challenges will require consolidated efforts
across the global signaling and drug discovery communities, but these
communities have lacked a common platform to integrate, overview
and analyze the mounting multifaceted data. Here, we provide an online
Biased Signaling Atlas enabling community collaboration and allowing
any researcher to tap into present data, address scientific questions and
deposit data for wider dissemination in the community hub.
Biased Signaling Atlas resources
The Biased Signaling Atlas comprises a collection of different data
and tool pages (summarized in Fig. 2). The atlas features interactive
resources to find reference ligands for bias experiments, quantify
Comment

ligand bias, navigate biased ligands, assess the experimental support Resource Data/tool visualization Questions
for bias, prioritize ligands to be tested for bias, explore pathway func-
Reference Which is, and where can I buy, the
tion and therapeutic relevance, and disseminate published results. ligand best reference ligand for my bias
selection experiments?
Guidelines and bias calculation. To aid those entering the field of
Template to How can I correctly
biased signaling, or seeking advice on how to design and report com- calculate bias calculate ligand bias?
plex ligand bias experiments, the first section, GUIDELINES, provides a
page ‘Community guidelines paper’, referencing the community guide- Biased ligand Which receptors and pathways
lines for GPCR ligand bias (above)11. A key consideration in study design coverage have biased ligands reported?
is to choose a reference ligand that can support the given scientific
Which biased ligands exist for my
question. Biased signaling in the context of physiology, a pathway or Biased
receptor pathway, and how strong
ligands
a compound of special interest requires a reference that is an endog- is their experimental evidence?
enous agonist, a pathway-balanced ligand or a user-selected drug or 15 How consistent is the relative bias
tool compound, respectively. Help in selecting the optimal reference of my ligand of interest across
Ligand bias 10
multiple studies?
ligand for bias is provided on a page entitled ‘Reference ligand selec- rank orders 5
How do different experimental setups
tion’ with links to relevant ligand resources. For the reporting, a key 0 or parameters affect the bias?
consideration is how to quantify ligand bias factors, as making correct Ligand What are the relative responses
calculations across ligands and pathways can be challenging. There- pathway of ligands across pathways and
profiles publications?
fore, the GUIDELINES section also provides an accessible spreadsheet
template to calculate ligand bias factors using predefined formulas and 3 What is the pathway preference of
a minimum set of pharmacological parameters—potency and efficacy, Pathway- 2 ligands not yet assessed against a
preferring reference ligand, and which of them
or signal transduction coefficient (values of Emax and EC50, or τ/KA, as ligands 1
should be prioritized for future bias
defined in Box 1 of ref. 11). 0 studies?

Which pathways have functional

Biased ligands. The BIASED LIGANDS (TRANSDUCER FAMILY/
SUBTYPE) sections present biased ligands for transducer families
(for example, Gi/o versus Gq/11) and subtypes (such as Gi1 versus Go),
respectively. These data and tools build on a unique dataset of 8,956
ligand-receptor-pathway activities curated from 212 literature reports.
This greatly exceeds the only previous resource for biased ligand data,
BiasDB17, which has 727 activities, is no longer maintained and lacks
information on bias type (above), quantitated bias factors, pharmaco-
logical parameters and key experimental details. The ligand bias data
are greatly expanded by incorporating publications that, although
not presented as biased signaling studies, contain the minimum infor-
mation to calculate bias: data describing a ligand with potencies and
efficacies at two pathways for the same receptor. To increase the com-
parability of results, all bias comparisons are restricted to the same cells
or similar cell lines derived from the same tissue, and to the same path-
way transducer/downstream level (http://docs.biasedsignalingatlas.
org). Furthermore, the atlas presents consistent bias factors using the
predominant quantification models: relative activity, log(Emax/EC50),
or operational model, log(τ/KA). Through the use of alternative refer-
ence ligands from the same study, the atlas uniquely covers different
types of bias: (i) pathway bias (versus pathway-balanced ligand), (ii)
physiology bias (versus endogenous agonist) and (iii) benchmark bias
(versus user-selected drug/tool compounds). This makes it possible to
carry out different studies investigating physiological, therapeutic and
experimental aspects, respectively, of biased signaling.
The ‘Biased ligand coverage’ page shows the distribution of
physiology-biased or pathway-biased ligands (bias factor >5) across
GPCRs and pathways. The top of the page displays biased ligand counts,
which are tabulations of the overall numbers of biased ligands and
GPCRs with a biased ligand, respectively, across the human GPCR
classes. Currently, this spans 180 pathway-biased ligands for 234 dis-
tinct receptors and 388 physiology-biased ligands for 223 GPCRs across
GPCR classes A, B1 and C. This is followed by a visualization of the biased
ligand coverage for each GPCR class and pathway as either a heat-
map or a circular tree, which further classifies receptors into families
Pathway mapping?
effects Which pathways have therapeutic
benefits and what is the basis?
How do I increase the exposure
when publishing ligand bias and
Data pathway effects?
deposition How do I compare my data to
those already stored in the biased
signaling atlas?
Fig. 2 | Resources and example research questions. The Biased Signaling Atlas
provides resources for retrieval of biased ligand and pathway effect data, analysis
of ligand-bias relationships, study design, bias calculation and data deposition.
sharing endogenous ligands (as a coloring option and tree branches,
respectively). To facilitate their use in publications and presentations,
these can be downloaded as images, with the option of interchanging
between receptor names from UniProt18 or the International Union of
Basic and Clinical Pharmacology (www.guidetopharmacology.org/
nomenclature.jsp).
The ‘Biased ligands’ browser presents ligands with quantita-
tive or qualitative bias values. To aid the selection of any receptor
of interest with a minimum number of biased ligands, the preced-
ing receptor selection page lists the number of pathway-biased and
physiology-biased ligands. Currently, the receptors with the largest
numbers of pathway-biased ligands (relative to a pathway-balanced
reference ligand) are the adenosine A3, δ-opioid, cannabinoid 1, dopa-
mine D2 and μ-opioid receptors (with 27, 25, 21, 11 and 8 such ligands,
respectively). Furthermore, the receptors with the largest numbers
of physiology-biased ligands (relative to an endogenous reference
ligand) are the μ-opioid, κ-opioid, dopamine D2, β2-adrenoceptor and
cannabinoid 2 receptors (with 43, 42, 40, 34 and 25 such ligands, respec-
tively). To aid researchers in selecting ligands for new studies based
on validation and availability, the ‘Biased ligands’ browser provides
reference and laboratory counts and vendors for tested biased ligands.
The browser then details consistent and structured bias information

nature chemical biology

Comment

a GPCR pathway-biased ligands b GPCR physiology-biased ligands

OPRD/δ DRD2/D2
GPR84 OPRM/µ
DRD2/D2 OPRK/κ
AA3R/A3 CNR2/CB2
CNR2/CB2 ADRB2/β2

CCR1/CCR1 HTR1A/5-HT1A

CNR1/CB1 HTR2C/5-HT2C

APJ/apelin OPRX/NOP

HTR2C/5-HT2C AGTR1/AT1
ACM5/M5 AA3R/A3
Total: 12 GPCRs 0 5 10 15 20 Total: 32 GPCRs 0 5 10 15 20 25 30 35 40

c GPCR pathway effect data points d

OPRM/µ Therapeutic area data points
AGTR1/AT1
315 data points
ADRB1/β1
Endocrine
DRD2/D2 Cardiovascular Psychiatric system
disease disorder disease
GLP1R/GLP1 53% 15% 2%
PAR2/PAR2
PAR1/PAR1 Analgesia/ Metabolic Nervous
pain therapy or nutritional system
APJ/apelin
22% disease disease
LPAR2/LPAR2 7% 1%

LPAR1/LPAR1
Total: 29 GPCRs 0 10 20 30 40 50
Fig. 3 | Top GPCRs and therapeutic areas. a–c, GPCRs with the largest numbers of pathway-biased ligands, physiology-biased ligands and pathway experiments,
respectively (bias factor cut-off >5). GPCR names: UniProt18/Guide to Pharmacology24. d, Therapeutic areas for annotated pathways.

(pathway rank orders, bias factors, specific pathways, operational
model bias and relative activity bias) together with the supporting
underlying pharmacological parameters (ligand Δlog(Emax/EC50) val-
ues, ligand Δlog(τ/KA) values, potency (pEC50) and Emax (% of reference
ligand)). Finally, it lists key experimental details (measured molecules,
biological process, cell lines and time resolution) with references to
the original papers.
‘Ligand bias rank orders’ visualizes the relative bias of ligands as a
scatter plot. Similarly, ‘Ligand pathway profiles’ visualizes the relative
pathway responses of ligands (without a reference ligand, and hence
not biased) as line charts or radar plots. Both tools support two comple-
mentary scientific use cases — comparison of different ligands from the
same study or of one ligand across publications — and image download.
Pathway-preferring ligands. The PATHWAY-PREFERRING LIGANDS
section describes each ligand’s activity across pathways and contains
the same four data and tool resources as the two ‘Biased ligands’ sec-
tions (above). However, ligand pathway preference differs from ligand
bias in using no reference ligand, and it therefore cannot distinguish
bias introduced by the ligand versus the system (biological, biosen-
sor, experimental setup, and so on)11. A common use case for ligand
pathway preferences is the selection of a reference ligand for ligand
bias studies. Here, the ligand browser has a special utility as it provides
information to reduce the impact of system bias by focusing on same or
near-identical systems and assays. Another advantage of this browser,
over the often-used mere fold potency measures, is that it presents
relative activities (Δlog(Emax/EC50)) accounting for differing efficacies,
which can substantially influence signaling responses. An alternative
use case is rank ordering of ligands when a fixed reference point cannot
be defined because a suitable reference ligand is lacking. This type of
ranking is possible through the use of the bias rank order and ligand
pathway profile tools.
Pathway effects. The PATHWAY EFFECTS section provides a functional
annotation of signaling pathways spanning physiological responses,
therapeutic effects or adverse effects. So far, the featured browser cov-
ers 638 pathway effects (outcomes or therapeutic areas) derived from
57 literature reports on diverse experiments, of which a subset have
used a ligand. These data may inform rational drug design targeting
‘on-pathways’ while avoiding ‘off-pathways’, analogous to the way that
‘on-target’ and ‘off-target’ proteins mediate therapeutic and adverse
effects, respectively. However, this represents only a fraction of known
G protein19 and arrestin couplings13 (GPCR kinases lack systematic
profiling) — necessitating new data deposition as more pathways gain
functional characterization (below).
Data deposition. Our annotation of pathway effects is the most com-
prehensive to date and covers a total of 29 GPCRs with pathway effects

nature chemical biology

Comment
(see the online atlas for details). Given that there are 108 GPCR with
approved drugs, at least 66 additional receptors in clinical trials and
224 non-olfactory GPCRs with broad untapped therapeutic poten-
tial9, it would be extremely valuable for the biased signaling com-
munity to expand the coverage (and database deposition) of pathway
effect outcomes and therapeutic areas. Via the DATA DEPOSITION
section, we invite all researchers to submit ligand-pathway-activity
and pathway-effect measurements via standardized spreadsheets.
Integration in a consolidated hub will help the field to jointly advance
the characterization of ligands and pathways across laboratories with
complementary techniques, such as measuring signaling at the trans-
ducer level, downstream or in vivo. It will also support the community in
better assessing the reproducibility of published studies and conduct-
ing more conclusive meta-analyses. Furthermore, it offers advantages
to authors and journals through consistent formatting of supplemen-
tary materials and increased publication exposure.
Top ten GPCRs and therapeutic areas across atlas
annotations
We analyzed our unique datasets of 8,956 ligand-receptor-pathway
activities and 319 pathway effects to obtain an overview of which GPCRs
and therapeutic areas have been most pursued in annotated literature
studies. We found that the top ten GPCRs with pathway-biased and
physiology-biased ligands (Fig. 3a,b) encompass 16 distinct receptors,
among which the adenosine A3, cannabinoid CB2, dopaminergic D2 and
serotonin 5-HT2C receptors have both types of biased ligands. These
receptors span nine receptor families as defined by shared endogenous
agonists — the acetylcholine (M5), adenosine (A3), adrenoceptor (β2),
apelin, angiotensin (AT1), cannabinoid (CB1, CB2), chemokine (CCR1,
NOP), dopamine (D2), opioid (δ, κ, μ) and serotonin (5-HT1A, 5-HT2C) fami-
lies — as well as GPR84, which is an ‘orphan’ receptor lacking a cognate
physiological ligand. The large diversity among the receptors with the
largest number of biased ligands supports the wide presence of biased
signaling throughout receptor-mediated physiological processes.
Of the ten GPCRs with most annotated pathway effects (Fig. 3c),
we found the highest number of experiments (n = 49) for the μ-opioid
receptor. This is due to the notion that biased drugs at this receptor may
mitigate the US ‘opioid crisis’ caused by wide prescription for pain treat-
ment and the often-lethal adverse effect of breathing suppression20.
This is followed by the angiotensin AT1, β1-adrenoceptor, dopamine D2,
glucagon-like peptide (GLP) 1, proteinase-activated receptor (PARs)
1 and 2, apelin, and lysophospholipid (LPA) 1 and 2 receptors. Of the
therapeutic areas, most data points fall within cardiovascular disease
(53%), followed by analgesia/pain therapy (22%) and psychiatric disor-
der (16%), whereas the four remaining areas comprise only 1%–7% of
data apiece (Fig. 3d). This exemplifies how GPCR pathways are being
targeted, or avoided, to develop biased drugs across therapeutic areas.
However, many more future studies would be needed to both extend
and test the biological rationale underlying the strategies to target
specific diseases using biased drugs.
All of these highly studied receptors are targets of approved drugs
except the apelin, LPA 1 and 2, CCR1 and GPR84 receptors, with the last
two having clinical agents in phase 2 trials9. The apelin receptor has no
clinical agents but has a key role in early development such as gastrula-
tion, blood vessel formation and heart morphogenesis21. The two LPA
receptors are activated by lysophosphatidic acid and thereby mediate
diverse cellular activities including cell migration, proliferation, inflam-
mation and tissue damage repair22. CCR1 affects stem cell proliferation,
and antagonists are currently being investigated in clinical trials for
nature chemical biology
rheumatoid arthritis and chronic obstructive pulmonary disease22.
GPR84 is an understudied receptor expressed on the surface of immune
cells that is currently being targeted by clinical agents in phase 2 trials
for idiopathic pulmonary fibrosis23. The strong focus on drug targets
underlines the interest in, and importance of, exploiting biased signal-
ing as a new mechanism for drug discovery.
Conclusion and future perspectives. These resources address mani-
fold scientific questions and use cases (Fig. 2) from scientists, teachers
and students across disciplines, including pharmacologists, molecular
biologists, cell biologists, clinicians, drug developers, medicinal and
computational chemists, bioinformaticians and data scientists. Thus,
we expect that the Biased Signaling Atlas will serve as a scientific and
learning platform enabling the global scientific community. We are
committed to the long-term development of the atlas, which will be
expanded further — for example, by providing receptor residues con-
stituting bias determinants aiding design of pathway-biased probes
and drugs. We invite all researchers to submit ligand-pathway-activity
and pathway-effect data upon original or review publication and to ask
journals to promote database deposition.
Data availability
All data are available in the online Biased Signaling Atlas (https://
BiasedSignalingAtlas.org) and GitHub (https://github.com/protwis/
gpcrdb_data). Documentation is available at http://docs.biasedsign-
alingatlas.org.
Code availability
All open-source code can be obtained from GitHub (https://github.com/
protwis/protwis) under the permissive Apache 2.0 License (https://
www.apache.org/licenses/LICENSE-2.0).
Jimmy Caroli1, Alibek Mamyrbekov1,3, Kasper Harpsøe 1,
Sahar Gardizi1,4, Linda Dörries2, Eshan Ghosh1,
Alexander S. Hauser 1, Albert J. Kooistra 1 &
David E. Gloriam 1
1
Department of Drug Design and Pharmacology, University of
Copenhagen, Copenhagen, Denmark. 2Department of Biochemistry,
Justus Liebig University Giessen, Giessen, Germany. 3Present address:
Trial Data Management, Novo Nordisk A/S, Søborg, Denmark.
4
Present address: IFP Manufacturing Development, Novo Nordisk A/S,
Bagsværd, Denmark.
e-mail: albert.kooistra@sund.ku.dk; david.gloriam@sund.ku.dk
Published online: xx xx xxxx
References
1. Kohout, T. A. et al. J. Biol. Chem. 279, 23214–23222 (2004).
2. Gomes, I. et al. Proc. Natl Acad. Sci. USA 117, 11820–11828 (2020).
3. Spengler, D. et al. Nature 365, 170–175 (1993).
4. Hollenberg, M. D. et al. Br. J. Pharmacol. 171, 1180–1194 (2014).
5. Schmid, C. L., Raehal, K. M. & Bohn, L. M. Proc. Natl Acad. Sci. USA 105, 1079–1084 (2008).
6. Dean, T., Vilardaga, J. P., Potts, J. T. Jr & Gardella, T. J. Mol. Endocrinol. 22, 156–166 (2008).
7. Smith, J. S., Lefkowitz, R. J. & Rajagopal, S. Nat. Rev. Drug Discov. 17, 243–260 (2018).
8. Kenakin, T. & Christopoulos, A. Nat. Rev. Drug Discov. 12, 205–216 (2013).
9. Hauser, A. S., Attwood, M. M., Rask-Andersen, M., Schioth, H. B. & Gloriam, D. E. Nat. Rev.
Drug Discov. 16, 829–842 (2017).
10. Foster, S. R. et al. Cell 179, 895–908 e821 (2019).
11. Kolb, P. et al. Br. J. Pharmacol 179, 3651–3674 (2022).
12. Anderson, A. et al. Proc. Natl Acad. Sci. USA 117, 14522–14531 (2020).
13. Avet, C. et al. Elife 11, 2020.2004.2020.052027 (2022).
14. Ghosh, E. et al. Cell Rep. 28, 3287–3299 e3286 (2019).
15. Wall, M. J. et al. Nat. Commun. 13, 4150 (2022).
Comment

16. Gillis, A. et al. Sci. Signal. 13, eaaz3140 (2020). COST Action ERNEST (CA18133), supported by COST (European Cooperation in Science and
17. Omieczynski, C. et al. Preprint at bioRxiv https://doi.org/10.1101/742643 (2019). Technology, www.cost.eu).
18. UniProt, C. Nucleic Acids Res. 49, D480–D489 (2021).
19. Hauser, A. S. et al. Elife 11, 22 (2022). Author contributions
20. Volkow, N. D. & Blanco, C. Mol. Psychiatry 26, 218–233 (2021). Conceptualization, D.E.G.; methodology, A.J.K. and D.E.G.; data curation, A.S.H., E.G., K.H., L.D.
21. Yang, P. et al. Circulation 135, 1160–1173 (2017). and S.G.; validation: A.J.K., D.E.G. and K.H.; writing: original draft, D.E.G.; writing: review and
22. Ochoa, D. et al. Nucleic Acids Res. 49, D1302–D1310 (2021). editing, all authors; visualization, D.E.G.; funding acquisition, D.E.G.; software, A.J.K., A.M. and
23. Wojciechowicz, M. L. & Ma'ayan, A. Nat. Rev. Drug Discov. 19, 374 (2020). J.C.; supervision, A.J.K., A.S.H, D.E.G. and K.H.
24. Harding, S. D. et al. Nucleic Acids Res. 50, D1282–D1294 (2022).
Competing interests
Acknowledgements D.E.G. and A.J.K. have part-time employments at Kvantify. After completing their contribution
This work was supported by grants from the Lundbeck Foundation (R313-2019-526) and the to this study, A.M. and S.G. moved to Novo Nordisk A/S. The remaining authors declare no
Novo Nordisk Foundation (NNF18OC0031226) to D.E.G. This article is based upon work from competing interests.

nature chemical biology