Structure–based discovery of opioid analgesics with reduced
side effects
Aashish Manglik
1,*
, Henry Lin
2,*
, Dipendra K. Aryal
3,*
, John D. McCorvy
3
, Daniela Dengler
4
,
Gregory Corder
5
, Anat Levit
2
, Ralf C. Kling
4,6
, Viachaslau Bernat
4
, Harald Hübner
4
, Xi-Ping
Huang
3
, Maria F. Sassano
3
, Patrick M. Giguère
3
, Stefan Löber
4
, Da Duan
2
, Grégory
Scherrer
1,5
, Brian K. Kobilka
1
, Peter Gmeiner
4
, Bryan L. Roth
3
, and Brian K. Shoichet
2
1
Department of Molecular and Cellular Physiology, Stanford University School of Medicine,
Stanford, California 94305, USA
2
Department of Pharmaceutical Chemistry, University of California, San Francisco, California
94158, USA
3
Department of Pharmacology, UNC Chapel Hill Medical School, Chapel Hill, North Carolina
27514, USA
4
Department of Chemistry and Pharmacy, Friedrich-Alexander-Universität Erlangen-Nürnberg,
Schuhstraße 19, 91052 Erlangen, Germany
5
Department of Anesthesiology, Perioperative and Pain Medicine, Neurosurgery, Stanford
Neurosciences Institute, Stanford University School of Medicine, Stanford, California 94305, USA
6
Institut für Physiologie und Pathophysiologie, Paracelsus Medical University, 90419 Nuremberg,
Germany
Reprints and permissions information is available at www.nature.com/reprints.
Correspondence and requests for materials should be addressed to B.K.K. (kobilka@stanford.edu), P.G. (peter.gmeiner@fau.de),
B.L.R. (bryan_roth@med.unc.edu) or B.K.S. (shoichet@cgl.ucsf.edu).
*
These authors contributed equally to this work.
Online Content Methods, along with any additional Extended Data display items and Source Data, are available in the online version
of the paper; references unique to these sections appear only in the online paper.
Supplementary Information is available in the online version of the paper.
Author Contributions A.M. and H.L. initiated the project. H.L. performed docking and identified compounds to be tested in the
initial and analogue screens. A.M. performed binding studies to identify initial hits and devised structure-guided optimization
strategies for subsequent analogues. D.K.A. performed
in vivo
studies, including analgesia assays, mouse plethysmography, faecal boli
accumulation studies, open field locomotor assay, and conditioned place preference. J.D.M., M.F.S. and P.M.G. performed radioligand
binding and signalling studies. X.P.H. performed signalling studies and assessed compound activity against the GPCRome. D.De.,
V.B., S.L. and H.H. synthesized compounds and determined affinities by radioligand binding and performed signalling studies. A.L.
and A.M. docked PZM21 and TRV130 and R.C.K. simulated PZM21 binding to μOR. G.C. performed reflexive and affective
analgesia studies of μOR knockout mice and was supervised by G.S. D.Du. performed pharmacokinetic studies. The manuscript was
written by A.M., H.L. and B.K.S. with editing and suggestions from B.L.R. and input from D.K.A., B.K.K. and P.G. P.G. supervised
chemical synthesis of compounds and the separation and identification of diastereomers, B.K.K. supervised testing of initial docking
hits, B.L.R. supervised radioligand binding, signalling and
in vivo
studies and B.K.S. supervised the compound discovery and design.
The project was conceived by A.M., H.L., B.K.K., P.G., B.K.S and B.L.R.
The authors declare competing financial interests: details are available in the online version of the paper. Readers are welcome to
comment on the online version of the paper.
Reviewer Information
Nature
thanks G. Henderson, E. Kelly, B. Kieffer and J. Meiler for their contribution to the peer review of this
work.
HHS Public Access
Author manuscript
Nature
. Author manuscript; available in PMC 2017 March 08.
Published in final edited form as:
Nature
. 2016 September 08; 537(7619): 185–190. doi:10.1038/nature19112.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Abstract
Morphine is an alkaloid from the opium poppy used to treat pain. The potentially lethal side
effects of morphine and related opioids—which include fatal respiratory depression—are thought
to be mediated by μ-opioid-receptor (μOR) signalling through the β-arrestin pathway or by actions
at other receptors. Conversely, G-protein μOR signalling is thought to confer analgesia. Here we
computationally dock over 3 million molecules against the μOR structure and identify new
scaffolds unrelated to known opioids. Structure-based optimization yields PZM21—a potent G
i
activator with exceptional selectivity for μOR and minimal β-arrestin-2 recruitment. Unlike
morphine, PZM21 is more efficacious for the affective component of analgesia versus the reflexive
component and is devoid of both respiratory depression and morphine-like reinforcing activity in
mice at equi-analgesic doses. PZM21 thus serves as both a probe to disentangle μOR signalling
and a therapeutic lead that is devoid of many of the side effects of current opioids.
Opiate addiction, compounded by the potentially lethal side effects of opiates such as
respiratory depression, has driven optimization campaigns for safer and more effective
analgesics since the 19th century. Although the natural products morphine and codeine, and
the semi-synthetic drug heroin, are more reliably effective analgesics than raw opium, they
retain its liabilities. The classification of opioid receptors into μ, δ, and κ and nociception
subtypes
1,2
raised hopes that subtype-specific molecules would lack the liabilities of
morphinan-based opiates. Despite the introduction of potent synthetic opioid agonists like
methadone and fentanyl, and the discovery of endogenous opioid peptides
3
, developing
analgesics without the drawbacks of classic opioids has remained an elusive goal. Recent
studies have suggested that opioid-induced analgesia results from μOR signalling through
the G protein G
i
, while many side effects, including respiratory depression and constipation,
may be conferred via β-arrestin pathway signalling downstream of μOR activation (Fig.
1a)
4–6
. Agonists specific to the μOR and biased towards the G
i
signalling pathway are
therefore sought both as therapeutic leads and as molecular probes to understand μOR
signalling. Recent progress has supported the feasibility and potential clinical utility of such
biased μOR agonists
7,8
.
The determination of the crystal structures of the μ, δ, κ and nociceptin opioid receptors
9–12
(Fig. 1b, c) provided an opportunity to seek new μOR agonists via structure-based
approaches. Recent discovery campaigns have used crystal structures of other Family A G-
protein-coupled receptors (GPCRs) to computationally dock large libraries of molecules,
identifying ligands with new scaffolds and with nanomolar-range potencies
13–17
. We thus
targeted the μOR for structure-based docking, seeking ligands with new chemotypes. We
reasoned that such new chemotypes might confer signalling properties with new biological
effects, as has been true for other structure-based campaigns
18,19
.
Structure-based docking to the µOR
We docked over 3 million commercially available lead-like compounds
20
against the
orthosteric pocket of inactive μOR
9
, prioritizing ligands that interact with known affinity-
determining residues and with putative specificity residues that differ among the four opioid
receptor subtypes (Fig. 1b, d). For each compound, an average of 1.3 million configurations
Manglik et al. Page 2
Nature
. Author manuscript; available in PMC 2017 March 08.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
was evaluated for complementarity to the receptor using the physics-based energy function
21
in DOCK3.6. As is common in docking
22,23
and screening, the top ranking molecules were
inspected for features not explicitly captured in the scoring function. We manually examined
the top 2,500 (0.08%) docked molecules for their novelty, their interactions with key polar
residues such as Asp147
3.32
(superscripts indicate Ballesteros-Weinstein numbering
24
), and
deprioritized those that showed conformational strain (a term occasionally poorly modelled
by the scoring function). Ultimately, 23 high-scoring molecules with ranks ranging from 237
to 2,095 out of the over 3 million docked were selected for testing (Fig. 1e). Compared to
the 5,215 μOR ligands annotated in ChEMBL16
25
, these docking hits had Extended
Connectivity Fingerprint 4 (ECFP4)-based Tanimoto coefficients (
T
c
) ranging from 0.28 to
0.31, which is consistent with the exploration of novel scaffolds
26
. Of the 23 tested, seven
had μOR binding affinities (
K
i
) ranging from 2.3 μM to 14 μM (Extended Data Table 1,
Extended Data Fig. 1).
The new ligands are predicted to engage the μOR in new ways (Fig. 1f and Extended Data
Fig. 1). Most opioid ligands use a cationic amine to ion-pair with Asp147
3.32
, a canonical
interaction
27
observed in structures of the μOR, δOR, κOR and nociceptin receptor bound to
ligands of different scaffolds
9–12,28
. As anticipated, the docked ligands recapitulated this
interaction. Much less precedence exists for the formation of an additional hydrogen bond
with this anchor aspartate, often mediated in the docking poses by a urea amide. In several
of the new ligands the urea carbonyl is modelled to hydrogen bond with Tyr148
3.33
, while
the rest of the ligands often occupy sites unexplored by morphinans (Extended Data Fig. 1).
To our knowledge, the double hydrogen bond coordination of Asp147
3.32
modelled in the
docking poses has not been anticipated or observed previously for opioid ligands, and only
50 of the 5,215 annotated opioid ligands in ChEMBL16 contain a urea group.
Despite the structural novelty of the initial docking hits, their affinities were low. To enhance
binding and selectivity, we docked 500 analogues of compounds 4, 5 and 7 that retained the
key recognition groups but added packing substituents or extended further towards the
extracellular side of the receptor, where the opioid receptors are more variable. Of the 15
top-scoring analogues that were tested, seven had
K
i
values between 42 nM and 4.7 μM
(Extended Data Table 2). Encouragingly, several were specific for the μOR over κOR
(compounds 12–15, Extended Data Table 2). We then investigated the more potent
analogues for signalling potency and efficacy. Although the structure we docked against was
the inactive state of the μOR, compounds 8 and 12–14 activated G
i/o
(Extended Data Table
2). A similar enrichment for agonists was previously seen in a docking study against the
inactive state of the κOR
23
, perhaps reflecting the small changes in the orthosteric pocket
associated with opioid receptor activation
29
. Encouragingly, the most potent compound, 12
(Fig. 2a, b), strongly activated G
i/o
with low levels of β-arrestin-2 recruitment (Fig. 2c, d).
Structure-guided synthetic optimization
To optimize compound 12, we synthesized stereochemically pure isomers and introduced a
phenolic hydroxyl (Fig. 3a). The synthesis of the (
S,S
) stereoisomer of 12 improved affinity
(
K
i
) to 4.8 nM and had a signalling EC
50
of 65 nM; it was the most potent and efficacious
G
i/o
signalling agonist among the four isomers (Fig. 3e). The phenolic hydroxyl, introduced
Manglik et al. Page 3
Nature
. Author manuscript; available in PMC 2017 March 08.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
to make compound (
S
,
S
)-21, was designed to exploit a water-mediated hydrogen bond with
His297
6.52
, an interaction observed in the structure of μOR in complex with β-
funaltrexamine (β-FNA) (Fig. 3b) and in other structures of the δOR
28
and κOR
11
. This
hydroxyl was readily accommodated in the docked μOR-12 complex, improving the
predicted docking energy (Fig. 3c). Compound (
S,S
)-21 had an EC
50
of 4.6 nM in a G
i/o
activation assay, with 76% efficacy (Fig. 3f), and a
K
i
of 1.1 nM in radioligand binding
assays (Extended Data Table 3), an improvement of 40-fold versus 12. The other three
stereoisomers of (
S
,
S
)-21 were much less potent or efficacious (Extended Data Fig. 2a, b),
suggesting a specific stereochemical requirement for both potency and efficacy in agreement
with the docked poses of (
S
,
S
)-21 to the inactive and active structures
29
of μOR (Fig. 3c,
Extended Data Fig. 2c, d). We refer to (
S
,
S
)-21 as compound PZM21 henceforth.
Because PZM21 was discovered against the inactive structure of μOR, its docked complex to
active μOR retains ambiguities. To investigate its receptor-bound structure further, more
detailed docking and molecular dynamics simulations were conducted. The resulting model
was tested by synthesizing molecules that either perturbed or exploited specific modelled
interactions (Fig. 3c, d, Extended Data Figs 2 and 3). Neutralization of charge by amidation
(compound PZM28) decreases potency by 1,000-fold, supporting a key ionic interaction
between the PZM21 tertiary amine and Asp147
3.32
(Fig. 3d and Extended Data Fig. 3).
Compound PZM27, which adds steric bulk to the tertiary amine, was synthesized to disrupt
putative hydrophobic interactions between the
N
-methyl group and Met151
3.36
and
Trp293
6.48
, consistent with its 30-fold loss of potency and decreased efficacy (Fig. 3d and
Extended Data Fig. 3). Compounds PZM23, PZM24 and PZM25, which were synthesized to
disrupt hydrogen bonding interactions in the model between the urea and Asp147
3.32
,
Tyr326
7.43
and Gln124
2.60
, lose between 30- and 230-fold potency despite their decreased
solvation penalties (Fig. 3d and Extended Data Fig. 3). These key ionic and hydrogen-
bonding interactions are maintained for 3 μs of molecular dynamics simulations of PZM21
in complex with active μOR, as are interactions between the phenolic hydroxyl and the
bridging waters to His297
6.52
, further supporting their relevance to the modelled pose
(Extended Data Fig. 2g). The thiophene of PZM21, modelled to fit in the more open
specificity region of the μOR, can be replaced with a larger benzothiophene without loss of
potency (Extended Data Fig. 3). Interactions of this thiophene with residues that differ
among the opioid receptor sub-types may contribute to PZM21 specificity (Extended Data
Fig. 2e). More compellingly, the simulations and docking predict that the PZM21 thiophene
comes within 6 Å of Asn127
2.63
in the active μOR (Extended Data Fig. 2g). Accordingly, we
synthesized an irreversible version of PZM21 (compound PZM29) designed to form a
covalent bond with μOR engineered with an N127C mutation. Compound PZM29 binds
irreversibly to this mutant but not the wild-type receptor and retains its efficacy as an agonist
(Extended Data Fig. 3), supporting the overall orientation of PZM21 as modelled and
simulated in the orthosteric μOR site.
PZM21 is a selective G
i
-biased µOR agonist
PZM21 had no detectable κOR or nociceptin receptor agonist activity—it is actually an 18
nM κOR antagonist—while it is a 500-fold weaker δOR agonist (Extended Data Fig. 4 and
Extended Data Table 3), making it a selective μOR agonist. To investigate specificity more
Manglik et al. Page 4
Nature
. Author manuscript; available in PMC 2017 March 08.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
broadly, PZM21 was counter-screened for agonism against 316 other GPCRs
30
. Activity at
10μM was observed at several peptide and protein receptors; however, no potent activity was
confirmed with a full dose–response experiment at these receptors. PZM21 therefore has
high agonist specificity among GPCRs (Extended Data Fig. 5a–c). PZM21 was also tested
for inhibition of the hERG ion channel and the dopamine, norepinephrine and serotonin
neurotransmitter transporters. At hERG, PZM21 had an IC
50
of between 2 and 4 μM, 500- to
1,000-fold weaker than its potency as a μOR agonist (Extended Data Fig. 5d). Its inhibition
of the neurotransmitter transporters, which are also analgesia targets, was even weaker with
IC
50
values ranging from 7.8 to 34 μM (Extended Data Fig. 5e). Thus, PZM21 is a potent,
selective, and efficacious μ opioid agonist.
A major goal of this study was to find new chemotypes that might display biased signalling
and perhaps, unlike canonical opioid drugs, have more favourable
in vivo
profiles. Signalling
by PZM21 and other μOR agonists appears to be mediated primarily by the heterotrimeric G
protein G
i/o
, as its effect on cAMP levels was eliminated by pertussis toxin and no activity
was observed in a calcium release assay (Extended Data Fig. 6a–d). A maximal
concentration of PZM21 led to no detectable β-arrestin-2 recruitment in the PathHunter
assay (DiscoverRx) (Fig. 3g and Extended Data Fig. 6c) and a minimal level of μOR
internalization compared to DAMGO and morphine (Extended Data Fig. 6e). Indeed, β-
arrestin-2 recruitment was too low to even permit a formal calculation of bias
31
, which
quantifies the preference for one signalling pathway over another. Since β-arrestin
recruitment can depend on the expression level of G protein-coupled receptor kinase 2
(GRK2)
32
, we also investigated G
i/o
signalling and arrestin recruitment in cells co-
transfected with this kinase. Even in the presence of overexpressed GRK2, PZM21 still has
weak arrestin recruitment efficacy compared to DAMGO and even to morphine (Extended
Data Fig. 6g–i). In fact, the signalling bias of PZM21 was undistinguishable from TRV130,
a G
i
-biased opioid agonist now in Phase III clinical trials (Fig. 3f, g), whereas its G-protein-
bias substantially exceeded that of herkinorin, which has also been purported to be a G
i
-
biased agonist
33
(Extended Data Fig. 6). An intriguing distinction in these signalling studies
is the lack of agonist activity of PZM21 at κOR. While PZM21 is an 18-nM antagonist of
this receptor, the other biased agonist, TRV130, activates κOR with similar potency to
morphine (Extended Data Fig. 6f). Additionally, despite having similar levels of signalling
bias, in modelling studies TRV130 and PZM21 appear to engage the μOR in distinct ways
(Extended Data Fig. 2f).
Analgesia with diminished side effects
Consistent with its μOR agonist activity, PZM21 displayed dose-dependent analgesia in a
mouse hotplate assay, with a per cent maximal possible effect (% MPE) of 87% reached 15
min after administration of the highest dose of drug tested (Fig. 4a). The highest dose of
morphine tested plateaued at 92% after 30 min. Intriguingly, we observed no analgesic effect
for PZM21 in the tail-flick assay (Fig. 4b). Such a distinction is unprecedented among
opioid analgesics. The hotplate experiment assesses analgesia at both higher-level central
nervous system (CNS) brain and spinal nociceptive circuits, while the tail-flick experiment
is more specific for spinal reflexive responses
34
. Subcategorizing the behavioural responses
in the hotplate experiment as either affective (CNS mediated) or reflexive (spinally
Manglik et al. Page 5
Nature
. Author manuscript; available in PMC 2017 March 08.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript