Structure-based discovery of opioid analgesics with reduced side effects

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.

Extended Data Figure 1. Docking poses of active compounds

Seven of 23 experimentally tested compounds bound to the μOR with micromolar affinity. Their docked poses often occupy sites not exploited by the antagonist β-funaltrexamine. In each case, a canonical ionic interaction with D147^3.32 is observed.

Extended Data Figure 2. Stereochemical structure-activity relationship

a, As with the different stereoisomers of 12, variation of the chiral centres in compound PZM21 results in large changes in efficacy and potency. Data are mean ± s.e.m. of normalized results (n = 3 measurements). b, Structure–activity relationship of compound 12 and 21 stereoisomers with affinities displayed as pKi values and agonist potency and efficacy in a G_i/o Glosensor assay. c, d, PZM21 docked to active μOR shows a more extended conformation as compared to the inactive state. e, In the docked active state, the PZM21 thiophene extends into the specificity-determining region of opioid receptors. Key interacting residues here are highlighted as red lines and corresponding residues at the other human opioid receptors are indicated. f, Docked pose of TRV130 within the μOR site, showing minimal overlap in key pharmacophores with PZM21 besides the ionic interaction between the cationic amine and D147^3.32. g, Molecular dynamics simulations of PZM21 in the inactive μOR state (grey and black traces) leads to a stable conformation with the thiophene positioned > 10 Å away from N127^2.63 (total of 2 μs of simulation time over three independent trajectories). In contrast, PZM21 adopts a more extended pose when simulated with active μOR, with an average distance of 6 Å between the thiophene and N127^2.63. Other key interactions between μOR and PZM21 are also highlighted.

Extended Data Figure 3. Structure activity relationship defined by PZM21 analogues

Eight analogues were synthesized to probe the binding orientation of PZM21 and their efficacy as agonists was tested in a CAMYEL-based G_i/o signalling assay. Analogues were compared to a parent reference compound (PZM22) with similar efficacy and potency to PZM21. In each case, the EC₅₀ value for PZM22 is shown in black (1.8 nM) and the EC₅₀ for the analogue is coloured. The covalent compound PZM29 binds to the μOR:N127C variant irreversibly, as evidenced by wash-resistant inhibition of radioligand binding. Signalling data are mean ± s.e.m. of normalized results (n = 3 measurements).

Extended Data Figure 4. Signalling properties of PZM21 at the opioid receptors

Displayed are raw luminescence data from a G_i/o Glosensor assay. In agonist mode, agonists decrease luminescence while inverse agonists increase it by diminishing basal signalling. For each opioid receptor, a prototypical well-characterized agonist (black curves) and antagonist (red curves) were used to validate the assay. In antagonist mode, a competition reaction is performed with 50 nM agonist and an escalating amount of tested drug. Here, true antagonists increase the observed signal, consistent with their ability to compete with the agonist but not induce G_i signalling. Data are mean ± s.e.m. of non-normalized results (n =3 measurements).

Extended Data Figure 5. PZM21 is selective for μOR

a, Compound PZM21 was screened against 320 non-olfactory GPCRs for agonism in the arrestin recruitment TANGO assay. Each point shows luminescence normalized to basal level at a given GPCR, with vertical lines indicating the standard error of the mean. b, GPCRs for which PZM21 induces an increase in signal twofold over background were further tested in full dose–response mode. Several potential targets (GPR110, MCHR1R, PTGER1) did not show dose-dependent increase in signal and probably represent screening false positives. CXCR7 and SSTR4 did show dose-dependent signals at high concentrations of PZM21, and were further tested in non-arrestin signalling assays. c, PZM21 does not show a dose-dependent change in cAMP inhibition in a G_i/o Glosensor assay measuring SSTR4 activation, indicating that the single elevated point in b is probably a false positive result. d, e, Inhibition assays of hERG (d) and the dopamine transporter (DAT), norepinephrine transporter (NET), and serotonin transporter (SERT) (e) show that PZM21 has weak inhibitory activity ranging from 2–34 μM at these targets. For a, data are mean ± s.e.m. of non-normalized results (n = 4 measurements). For b–e, data are mean ± s.e.m. of normalized results (n = 3–6 measurements).

Extended Data Figure 6. Signalling profile of PZM21 and other μOR agonists

a, PZM21 is an efficacious G_i and G_o agonist in a GTP γS assay. b, Like other μOR agonists, PZM21 induces a dose-dependent decrease in cAMP levels that is sensitive to pertussis toxin, confirming G_i/o mediated signalling. c, Herkinorin is a G_i/o agonist and robustly recruits arrestin in a BRET assay performed in the absence of GRK2 overexpression. TRV130 or PZM21 show undetectable levels of arrestin recruitment in the same experiement. d, PZM21 and other opioids show no activity in a calcium-release assay, indicating no G_q-mediated second messenger signalling. The positive control TFLLR-NH2 efficiently activates the G_q coupled receptor PAR-1. e, PZM21 and TRV130 induce much decreased receptor internalization versus DAMGO and even morphine. f, Herkinorin and TRV130 are potent agonists of the κOR. PZM21 is a κOR antagonist (see Extended Data Fig. 4). g, In HEK293 cells, GRK2 expression levels have minimal effect on the potency and efficacy of the unbiased agonist DAMGO in a G_i/o activation assay. Increased GRK2 levels change the basal cAMP signal due to increased desentization of μOR, which lowers receptor basal activity and leads to elevated isoproterenol-induced cAMP. In an arrestin-recruitment BRET assay, increased GRK2 expression increases both the potency and maximal efficacy of the unbiased agonist DAMGO. This is likely because GRK2 mediated phosphorylation is required for efficient β-arrestin recruitment. h, G_i activation and arrestin recruitment in cells co-expressing 1.0 μg/15 cm² of GRK2. Notably, PZM21 induces a higher maximal level of arrestin recruitment as compared to U2OS cells, which express very low levels of GRK2, but this level is significantly lower than morphine. Despite the lower efficacy for arrestin recruitment observed for morphine, TRV130 and PZM21 compared to DAMGO, a formal calculation of bias by the operational models fails to show that this effect is significant. i, Table of pEC₅₀ values and E_max values for various signalling assays. All data are mean ± s.e.m. of results (n = 2–6 measurements).

Extended Data Figure 7. Additional in vivo studies of PZM21

a, Analgesic responses measured in the hotplate assay were subcategorized into either affective or reflexive behaviours and scored separately. b, Morphine (n = 10 animals) induces changes in both behaviours, while PZM21 (n = 13 animals) only modulates the attending (affective) component. Knockout of the μOR ablates all analgesic responses by morphine and PZM21. c, PZM21 shows minimal cataleptic effect compared to morphine at different time points. The effect of haloperidol was included as a positive control. d, Pharmacokinetic studies of PZM21 (n = 3–4 animals for each time point) show central nervous system penetration of the compound, with a peak level of 197 ng of PZM21 per g of brain tissue. With a concomitant serum concentration of 1,253 ng/ml, this represents a serum:brain concentration ratio of 6.4. These levels are similar to those achieved by morphine, which shows a peak brain concentration of approximately 300 ng/g and a serum:brain concentration ratio of 3.7 30 min after subcutaneous injection. e, Metabolism of PZM21 over 60 min exposure to mouse liver microsomes. Rotigotine and imipramine serve as positive controls for extensive phase I metabolism. The total amount of PZM21 and metabolite pool is slightly greater than 100% (101.8%) reflecting cumulative error in LC/MS analysis. f, A G_i/o signalling assay shows that none of the metabolites are measurably more potent activators of the μOR versus PZM21 alone. The metabolite pool after the 60-min incubation was used directly in the signalling assay. As a negative control, the pooled material from a reaction carried out in the absence of the key cofactor NADPH was used in the signalling assay. All data are mean ±s.e.m. For e, reactions were run in triplicate and the s.e.m. was calculated from individual measurements of each reaction.

Figure 1. Structure based ligand discovery for the μOR

a, Opiate-induced μOR signalling through G_i activates G-protein-gated inwardly rectifying potassium channels (GIRKs) and inhibits adenylyl cyclase, leading to analgesia. Conversely, recruitment of β-arrestin is implicated in tolerance, respiratory depression, and constipation. b, Cutaway of the μOR orthosteric site to which β-FNA binds. Highlighted regions on the extracellular side diverge between the opioid receptors. c, Conserved features of opioid ligand recognition in the μOR. d, Overlaid docking poses of 23 compounds selected for experimental testing. e, Single-point competition binding assay of 23 candidate molecules against the μOR antagonist ³H-diprenorphine. Each ligand was tested at 20 μM and for those with > 25% inhibition affinity was calculated in full displacement curves; data represent mean ± s.e.m. (n = 3 measurements). One of these hits, compound 7, was subsequently optimized. f, Docking pose of compound 7.

Figure 2. Discovery of a novel G_i/o-biased μOR agonist

a, Compound 12 was identified among a series of analogues to compound 7 and further investigated due to its μOR specificity and efficacy as a μOR agonist. b, Docking pose of Compound 12. c, Compound 12 is a μOR agonist in a G_i/o signalling assay with an EC₅₀ of 180 nM. DAMGO is a prototypical unbiased opioid agonist. d, Despite robust activation of G_i/o, compound 12 induces minimal arrestin recruitment as compared to DAMGO. For c, d, data are mean ± s.e.m. of normalized results (n = 3–6 measurements).

Figure 3. Structure-guided optimization towards a potent biased μOR agonist

a, Structure guided optimization towards PZM21. b, Docking pose of 12 compared to β-FNA. The phenolic hydroxyl of β-FNA coordinates His297^6.52 with two water molecules, providing an optimization strategy for 12. c, PZM21 docked to active μOR with a water-mediated network between the PZM21 phenol and His297^6.52. The co-crystallized agonist BU72 is shown as orange sticks. d, μOR–PZM21 interactions include hydrogen bonds (blue dash), hydrophobic interactions (green dash), and an ionic bond (red dash). Insets show select data from structure–activity relationship and molecular dynamics studies presented in more detail in Extended Data. e, Stereoisomers of 12 in a G_i/o signalling assay. f, G_i/o signalling assay shows robust μOR agonist activity for PZM21. g, PZM21 shows undetectable β-arrestin-2 recruitment in the PathHunter assay. For e–g, data are mean ± s.e.m. of normalized results (n = 3–6 measurements).

Figure 4. PZM21 is an analgesic with reduced on-target liabilities

a, Analgesia in the mouse hotplate assay. Latency of withdrawal to noxious stimuli is shown as percentage of the maximal possible effect (% MPE). The highest dose of PZM21 (40 mg kg⁻¹) yields an equi-analgesic response to 10 mg kg⁻¹ morphine and 1.2 mg kg⁻¹ TRV130 at 15 min. b, Compared to morphine, PZM21 shows no analgesia in the tail-flick assay. c, Unlike morphine, PZM21 decreases affective pain perception with minimal effect on reflexive pain. d, PZM21 and morphine produce sustained analgesia in a formalin injection nociception assay e, PZM21 shows no analgesic effect in Oprm1^−/− mice, supporting engagement of μOR in vivo. f, Constipatory effects of morphine and PZM21 compared to vehicle assessed by accumulated faecal boli. g, Whole-body mouse plethysmography shows decrease in respiratory frequency for morphine starting 20 min after administration of drug. An equi-analgesic dose of PZM21 has no effect on respiration versus vehicle, while TRV130 induces transient respiratory depression at 15 min. h, Open field locomotor assay. i, Place preference is induced by conditioning with 10 mg kg⁻¹ morphine but not with 20 mg kg⁻¹ PZM21 nor with 1.2 mg kg⁻¹ TRV130. Per cent of time spent in either vehicle or drug chamber before (pretest) or after (test) conditioning regimen. All data are mean ± s.e.m. and asterisks indicate statistically significant differences between drug and vehicle. The number of animals in each group and statistical tests are described in the Methods.