The Intriguing Effects of Substituents in the N-Phenethyl Moiety of Norhydromorphone: A Bifunctional Opioid from a Set of “Tail Wags Dog” Experiments

(−)-N-Phenethyl analogs of optically pure N-norhydromorphone were synthesized and pharmacologically evaluated in several in vitro assays (opioid receptor binding, stimulation of [35S]GTPγS binding, forskolin-induced cAMP accumulation assay, and MOR-mediated β-arrestin recruitment assays). “Body” and “tail” interactions with opioid receptors (a subset of Portoghese’s message-address theory) were used for molecular modeling and simulations, where the “address” can be considered the “body” of the hydromorphone molecule and the “message” delivered by the substituent (tail) on the aromatic ring of the N-phenethyl moiety. One compound, N-p-chloro-phenethynorhydromorphone ((7aR,12bS)-3-(4-chlorophenethyl)-9-hydroxy-2,3,4,4a,5,6-hexahydro-1H-4,12-methanobenzofuro[3,2-e]isoquinolin-7(7aH)-one, 2i), was found to have nanomolar binding affinity at MOR and DOR. It was a potent partial agonist at MOR and a full potent agonist at DOR with a δ/μ potency ratio of 1.2 in the ([35S]GTPγS) assay. Bifunctional opioids that interact with MOR and DOR, the latter as agonists or antagonists, have been reported to have fewer side-effects than MOR agonists. The p-chlorophenethyl compound 2i was evaluated for its effect on respiration in both mice and squirrel monkeys. Compound 2i did not depress respiration (using normal air) in mice or squirrel monkeys. However, under conditions of hypercapnia (using air mixed with 5% CO2), respiration was depressed in squirrel monkeys.


Introduction
It is well-known that various N-substituents in the classical opioid-type of structure (e.g., 4,5epoxymorphinans, morphinans, 6,7-benzomorphans, 5-phenylmorphans) can change prototypical N-methyl substituted agonist opioids to opioid antagonists. Most familiar, replacement of the Nmethyl with an N-allyl moiety in morphine and oxymorphone converted them to the antagonists nalorphine and naloxone (Scheme 1).

Scheme 1. Prototypical 4,5-epoxymorphinan agonists and antagonists.
Exactly how the N-substituent interaction with amino acid residues in the receptor induces that change remains uncertain. The N-methyl substituent in morphine and oxymorphone either permits or does not prevent the ligand from receptor interactions that result in analgesia and the well-known panoply of opioid side-effects. Concurrent with their interaction with G-proteins, potent clinically utilized opioids also recruit β-arrestin2. This recruitment has been hypothesized to underly the undesirable effects of opioids, including respiratory depression, inhibited gastrointestinal transport, and tolerance [1][2][3][4], although recent data using β-arrestin2 knockout mice cast doubt on that hypothesis [4].
Replacement of an N-methyl group with a different substituent, the N-phenethyl moiety, has also been shown to change an opioid agonist to an antagonist in at least one of the types of classical opioids; however, that outcome is not consistent for all opioid structures. Most N-phenethylsubstituted opioids, such as N-phenethylnormorphine [5], N-phenethylnoroxymorphone [6], and 2′hydroxy-5,9-dimethyl-N-phenethylbenzomorphan [6] have been found to be potent μ-opioid receptor (MOR) agonists, but N-phenethyl-5-phenylmorphan acts as a MOR antagonist [7]. In that 5-Scheme 1. Prototypical 4,5-epoxymorphinan agonists and antagonists.
Exactly how the N-substituent interaction with amino acid residues in the receptor induces that change remains uncertain. The N-methyl substituent in morphine and oxymorphone either permits or does not prevent the ligand from receptor interactions that result in analgesia and the well-known panoply of opioid side-effects. Concurrent with their interaction with G-proteins, potent clinically utilized opioids also recruit β-arrestin2. This recruitment has been hypothesized to underly the undesirable effects of opioids, including respiratory depression, inhibited gastrointestinal transport, and tolerance [1][2][3][4], although recent data using β-arrestin2 knockout mice cast doubt on that hypothesis [4].
Replacement of an N-methyl group with a different substituent, the N-phenethyl moiety, has also been shown to change an opioid agonist to an antagonist in at least one of the types of classical opioids; however, that outcome is not consistent for all opioid structures. Most N-phenethyl-substituted opioids, such as N-phenethylnormorphine [5], N-phenethylnoroxymorphone [6], and 2 -hydroxy-5,9-dimethyl-N-phenethylbenzomorphan [6] have been found to be potent µ-opioid receptor (MOR) agonists, but N-phenethyl-5-phenylmorphan acts as a MOR antagonist [7]. In that 5-phenylmorphan series, we found that the agonist vs antagonist activity of the compound was dependent on chirality.
Opioid receptor binding data (Ki, nM) and stimulation of ([ 35 S]GTPγS) binding were obtained for all of these compounds (Table 1). Two additional compounds were added to group 4 in Table 2 (forskolin-induced cAMP accumulation assay), a 4-bromo (2j) and a 2,6-dichlorophenethyl compound (2k), and two additional compounds were added to group 1, hydromorphone (S5) and N-phenethylnor hydromorphone (S11). These four compounds were only examined using the forskolin-induced cAMP accumulation assay. The ability of all of the compounds to recruit β-arrestin was also determined ( Table 2).     NT NT a Binding assays were typically conducted in at least three independent experiments, each performed with triplicate observations using whole rat brains excluding cerebellum; Ki ± SEM (nM); NT = not tested-inactive (<50% activity at 100 nM concentration in exploratory binding assays (displaced less than half of radioligand). Also, compounds with low binding affinity (>50 nM) were not further examined in functional assays); b [ 35       Standards To determine % efficacy of MOR-mediated β-arrestin recruitment, the highest dose(s) of DAMGO were used as 100% and the respective data were converted to percentages based on the response of DAMGO and analyzed using GraphPad Prism. To determine % efficacy in forskolin-induced cAMP assays, data were normalized to the vehicle control, followed by the forskolin control. Data were then analyzed in GraphPad Prism using nonlinear regression.

Opioid Receptor Binding, Ligand Efficacy and Potency ([ 35 S]GTPγS Binding Assays)
The MOR binding affinity, in the receptor binding assay, of N-cyanomethyl (1a) and N-cyanopropyl compounds (1b) in group 1 (Table 1), both without an aromatic ring on the nitrogen atom, were comparable to the affinity of morphine (Ki = 4.51 and 4.25 vs 3.26 for morphine). However, 1a had only partial MOR agonist activity (47% stimulation) and very low potency (EC 50 = 425 nM) in the [ 35 S]GTPγS assay (Table 1), and 1b, in contrast, was a MOR antagonist in that assay (antagonist activity was assumed given the high MOR binding affinity, and lack of [ 35 S]GTPγS stimulation at MOR). All of the compounds in group 1 ( Table 1) that were assessed at DOR had relatively low receptor binding affinity (Ki > 70 nM). At KOR, 1a had a Ki of about 90 nM, whereas 1b had a higher binding affinity at KOR than MOR, a > 30-fold increase in KOR agonist affinity due to the extension of the carbon chain from N-cyanomethyl to N-cyanopropyl. Compounds in group 1 with an extended and rotationally restricted N-substituent (compound 1c) or a bulky N-substituent (compound 1d) showed little binding affinity or potency at any opioid receptor. Compounds were considered inactive and were not tested (NT) in functional assays when they were found to displace <50% of the radioligand at a 100 nM concentration in an exploratory binding assay (not described in the Material and Methods Section).
The m-(2b) and p-N-nitrophenethylnorhydromorphone (2a) in group 2 in Table 1 showed a >10-fold difference in MOR binding affinity in the receptor binding assay and a remarkable change from potent MOR agonist 2a to MOR antagonist 2b (Table 1) was observed in the [ 35 S]GTPγS assay, apparently induced by the change in the position of the aromatic ring's substituent. This bolstered our hypothesis that we could influence and considerably alter activity by substitution on the phenyl ring.
The group 3 alkyl and alkoxy compounds had very high MOR affinity (Ki < 1 nM) and high DOR affinity (Ki = 5-6 nM) in the receptor binding assay, and while 2c was a potent partial MOR agonist in the [ 35 S]GTPγS assay, the methoxy compound 2d appeared to be a MOR antagonist in that assay. Both of these compounds had EC 50 < 35 nM at DOR with 2c acting as a full agonist (95% stimulation) and 2d a partial agonist (49% stimulation). Compound 2d was also a potent KOR agonist (EC50 = 5.9 nM), although it was not efficacious at KOR (21.8% stimulation).
The halides in group 4 (Table 1) harbored the most interesting compound 2i, from the perspective of having a desirable δ/µ potency ratio. All of the halides had high affinity at MOR and DOR (Ki ranged from 0.3 to 2.7 nM at MOR and 4 to 16 nM at DOR), and less affinity at KOR (Ki > 20 nM), in the receptor binding assays. Additionally, all group 4 compounds had nanomolar MOR potency in the [ 35 S]GTPγS assay (EC50 = 2.0-3.4 nM) and all except 2i and 2h had lower DOR agonist potency (EC50 > 50 nM). The trifluoromethyl compound 2h had moderate DOR potency (EC 50 = 36 nM), whereas 2i had nanomolar potency at DOR (EC50 = 2.4 nM), with a δ/µ potency ratio of 1.2.

Ligand Potency and Efficacy Using the Forskolin-induced cAMP Accumulation Assay
As seen in the forskolin-induced cAMP accumulation assay (group 1, Table 2), 1a had morphine-like potency, as it did in the [ 35 S]GTPγS assay. In contrast, compounds 1a, 1c and 1d had relatively low potency for DOR or KOR cAMP stimulation.
Again, as in the [ 35 S]GTPγS assay, 1c with restricted rotation and 1d with a bulky side-chain were less potent than the cyanomethyl compound 1a. The standard compounds for comparison purposes, S5 and S11, hydromophone and N-phenethylnorhydromorphone in group 1, were relatively potent at MOR (EC 50 = 1.67 and 0.04 nM, respectively) and S11 was potent at DOR (EC 50 = 1.54 nM) and somewhat less potent at KOR (EC 50 = 22.7 nM). The parent compound hydromorphone (S5) was essentially inactive at DOR and KOR (EC 50 > 130 nM). The N-phenethylnorhydromorphone (S11), had an EC 50 δ/µ ratio = 38.5, and that was possibly too high for the mitigation of side-effects that might be provided by interaction with DOR.
The bias factor of all of the tested compounds ranged between ca. 0.4 and 3.5 ( Table 2). Examination of the MOR β-arrestin recruitment for S11 and 2i indicated that they both had a lower bias factor than morphine (indicating their greater ability to recruit β-arrestin). All of these G-protein biased ligands recruit β-arrestin almost as well, or better than morphine. If recruitment of β-arrestin correlated with the side-effects of these compounds, they should all cause, for example, respiratory depression, as well as other side-effects caused by interaction with MOR. As noted previously, however, the ability of MOR ligands to recruit β-arrestin may not have any bearing on whether they will or will not display opioid-like side-effects [4].
The MOR antagonist profile of 2a in group 2 (Table 1), the p-nitro compound, that was seen in the [ 35 S]GTPγS assay was not observed in the forskolin-induced cAMP accumulation assay (Table 2). In the cAMP assay, both 2a and 2b had MOR agonist activity (EC 50 = 0.05 and 5.2 nM, respectively); a >100-fold potency change due to a positional shift of an aromatic substituent.
In group 3 in Table 2, the alkyl and alkoxy compounds, 2c and 2d, were found to have relatively high potency (EC 50 = 0.08 and 0.13 nM, respectively) at MOR and at DOR (EC 50 = 1.0 and 2.7 nM, respectively. They were also KOR agonists (EC 50 = 8.7 and 7.4 nM, respectively). In the cAMP assay (Table 2), 2d was not found to have MOR antagonist activity.
In group 4 (Table 2), the most interesting compound was again found to be 2i. It had extremely high agonist potency at MOR and DOR (EC 50 = 0.05 and 0.53 nM, respectively), and was efficacious at both receptors. It was much less potent at KOR (EC 50 = 55 nM). The o-fluoro compound 2g was notable for its relatively high MOR agonist potency (0.01 nM), with full (101%) efficacy. The p-bromo compound 2j also appeared to be of interest in that its EC 50 δ/µ ratio = 6.5 was less than the ratio found for 2i in the cAMP assay (δ/µ ratio = 10 for 2i). Compound 2j was also potent at both MOR and DOR in the cAMP assay (EC 50 < 1 nM), but it had moderate agonist potency at KOR (20 nM). Although KOR agonists have therapeutic potential, they also have undesirable CNS-mediated side-effects (e.g., dysphoria, hallucinations) [32].
We hypothesized that a compound with a δ/µ potency ratio of less than 7 in the [ 35 S]GTPγS assay would be desirable if the compound were to display other than the full array of undesirable effects that clinically used analgesics manifest. Recent work on bifunctional compounds explored compounds with δ/µ potency ratios of 5 to 7 in [ 35 S]GTPγS assay assays, and found that those compounds had a less disruptive effect on locomotor activity than morphine or oxymorphone [33].

Molecular Modeling and Simulations.
Specifically placed moieties at the tail end of the N-phenethyl substituent changed the compound from a MOR partial agonist in the [ 35 S]GTPγS assay (e.g., N-p-nitrophenethyl-norhydromorphone, 2a) to a MOR antagonist (e.g., N-m-nitrophenethylnorhydromorphone, 2b). In the forskolin-induced cAMP accumulation assay, that same positional shift of the substituent in the aromatic ring in 2a to 2b caused a >100 fold change in potency. We used quantum chemical calculations and molecular dynamics simulations to determine if a moiety in precise positions on the aromatic ring in the N-phenethyl moiety of norhydromorphone would display sufficient differences in their interaction with MOR to induce the change in activity and/or potency. More generally, our simulations allowed us to identify the critical residues interacting with the body (Figure 1) and the tail (Figure 2) of the ligands that are responsible for the differences in receptor properties. Together, the experimental and simulation data led us to propose a set of general rules for the N-phenethyl-substitutions to impart specific behaviors, such as partial or full agonist or possible antagonist activities, which may help design compound with novel properties (see details in the Supplementary Materials).

Body-Opioid Receptor (OR) interactions
All the compounds considered in our simulations have similar body-OR interaction patterns regardless of the substituents on the N-phenethyl ring; the interactions are consistent with those reported recently for a series of phenethyl oxymorphone compound bound to the active MOR [33]. These are shown in Figure 1 for both the MOR and the DOR, and involve polar/charged residues in transmembrane helix (TMH) 3, 5 and 6, and hydrophobic residues in TMH 5, 6, and 7. A contact is deemed significant if it persists for at least 50% of the time (see example in Table S2), although not necessarily all the interactions are seen at a given time. Different substituents, however, lead to different statistics of the individual interactions due to modest repositioning of the ligands resulting from different tail-OR interactions. All the residues in direct contact with the body are conserved in both receptors. Three additional, non-conserved residues, each belonging to TMH 5, 6, and 7 interact indirectly with -O-and -OH groups of the body via short water chains. Thus, suitable substitutions that engage these residues more directly (e.g., polar or H-bond interactions) may help modulate the MOR and DOR activity, potency and affinities independently.
All the compounds considered in our simulations have similar body-OR interaction patterns regardless of the substituents on the N-phenethyl ring; the interactions are consistent with those reported recently for a series of phenethyl oxymorphone compound bound to the active MOR [33]. These are shown in Figure 1 for both the MOR and the DOR, and involve polar/charged residues in transmembrane helix (TMH) 3, 5 and 6, and hydrophobic residues in TMH 5, 6, and 7. A contact is deemed significant if it persists for at least 50% of the time (see example in Table S2), although not necessarily all the interactions are seen at a given time. Different substituents, however, lead to different statistics of the individual interactions due to modest repositioning of the ligands resulting from different tail-OR interactions. All the residues in direct contact with the body are conserved in both receptors. Three additional, non-conserved residues, each belonging to TMH 5, 6, and 7 interact indirectly with -O-and -OH groups of the body via short water chains. Thus, suitable substitutions that engage these residues more directly (e.g., polar or H-bond interactions) may help modulate the MOR and DOR activity, potency and affinities independently. Although not in direct interactions, a few non-conserved residues (green) were seen to interact with the ligands indirectly through short water chains. These residues may thus be important to modulate the behavior of MOR and DOR independently, which may be accomplished through specific substituents that can engage them more directly. Frequencies of contacts were deemed statistically relevant if they were observed at least 25% of the time, except for the critical -NH---D147 distance (MOR) and -NH---D128 distance (DOR) that was required to persist for at least 75% of the time for the conformer to be considered (see details of the analysis in Table S2).

Tail-OR interactions
Because the tail is located deep inside the pocket, relatively small changes in the N-phenylethyl ring via substitutions perturb residues located in different regions of the ORs and engage different TMHs ( Figure S1). The constrained environment of the tail suggests that these interactions can induce major changes in the OR structure and/or dynamics, as confirmed by preliminary data from principal component analysis (to be published). Unlike the relatively rigid ligand body, the tail has several Figure 1. Conserved polar/charged residues (in red; numbering as in MOR) that interacted with the -O and -OH groups of the N-phenethylnorhydromorphone body through H-bonds; the body was also stabilized by close packing with four conserved nonpolar residues (blue) through hydrophobic forces. Although not in direct interactions, a few non-conserved residues (green) were seen to interact with the ligands indirectly through short water chains. These residues may thus be important to modulate the behavior of MOR and DOR independently, which may be accomplished through specific substituents that can engage them more directly. Frequencies of contacts were deemed statistically relevant if they were observed at least 25% of the time, except for the critical -NH-D147 distance (MOR) and -NH-D128 distance (DOR) that was required to persist for at least 75% of the time for the conformer to be considered (see details of the analysis in Table S2).

Tail-OR interactions
Because the tail is located deep inside the pocket, relatively small changes in the N-phenylethyl ring via substitutions perturb residues located in different regions of the ORs and engage different TMHs ( Figure S1). The constrained environment of the tail suggests that these interactions can induce major changes in the OR structure and/or dynamics, as confirmed by preliminary data from principal component analysis (to be published). Unlike the relatively rigid ligand body, the tail has several energetically similar conformers ( Figure S2). Some of these conformers can be ruled out based on unfavorable steric interactions (cf. Computational Section 5). However, the simulations show that all of the conformers selected by the pocket can be stabilized because each substituent can always find favorable interactions through polar/nonpolar or H-bond interactions. It is noted that none of these conformers lose the critical polar/H-bond interaction between the protonated nitrogen of the ligands and the carboxylate of the anchoring aspartic acid D147 (MOR) or D128 (DOR) (cf. Table S2). The multiplicity of binding modes may explain, in part, the tendency of most ligands in the Table 1 series to display partial agonist activity, i.e., with some modes leading to agonist and others to antagonist activity. This scenario may coexist with the more traditional view of partial activity as the result of multiple substates of the receptor stabilized by a single conformer (not observed in the simulations). We carried out a comparative analysis of the experimental observations summarized in Table S1. Here we focused on the results pertaining to the effects of F and Cl substitution at the p-position of the N-phenethyl moiety on both receptors (Figure 2A, additional details in SI). The p-F (2e) was found to be a potent partial MOR agonist and weak partial DOR agonist, whereas p-Cl (2i), although still a potent partial MOR agonist, becomes a potent full DOR agonist, an unexpected result with therapeutic potential.
conformers lose the critical polar/H-bond interaction between the protonated nitrogen of the ligands and the carboxylate of the anchoring aspartic acid D147 (MOR) or D128 (DOR) (cf. Table S2). The multiplicity of binding modes may explain, in part, the tendency of most ligands in the Table 1 series to display partial agonist activity, i.e., with some modes leading to agonist and others to antagonist activity. This scenario may coexist with the more traditional view of partial activity as the result of multiple substates of the receptor stabilized by a single conformer (not observed in the simulations). We carried out a comparative analysis of the experimental observations summarized in Table S1. Here we focused on the results pertaining to the effects of F and Cl substitution at the p-position of the N-phenethyl moiety on both receptors (Figure 2A, additional details in SI). The p-F (2e) was found to be a potent partial MOR agonist and weak partial DOR agonist, whereas p-Cl (2i), although still a potent partial MOR agonist, becomes a potent full DOR agonist, an unexpected result with therapeutic potential. One of the conformers of 2e and of 2i showed the same pattern of interactions with the DOR, engaging both TMH 6 (W274) and 7 (N310, S311, N314); this conformer (not shown) was expected to One of the conformers of 2e and of 2i showed the same pattern of interactions with the DOR, engaging both TMH 6 (W274) and 7 (N310, S311, N314); this conformer (not shown) was expected to elicit the same dynamic behavior of the receptor and was unlikely to be responsible for the observed partial vs. full agonist activities. The second conformer did show qualitative differences ( Figure S2A); when compared to 2e, 2i partially disengaged TMH 7 and engaged TMH 3. On the one hand, Cl is larger than F, resulting in higher polarizability together with a longer C-Cl bond distance, and is less electronegative, resulting in a less negative partial charge and weaker electric field at the atom surface. These differences appeared to be enough for 2i to interact more favorably with both S135 and S311, which were in opposite sides across the pocket; 2e instead interacted more closely with the polar groups of adjacent N310 and N314 on the same TMH 7. In both cases, the ligands interacted with TMH 6, showing persistent interactions with the -NH group of W274 and with the H atoms on the F270 ring via weak polar interactions. Although halogen-bonding interactions were not included in the forcefield [34,35], it could be predicted that the incipient σ-hole and equatorial negative charge of the Cl atom ( Figure S2) would further stabilize this pattern of interactions. There may be other qualitative differences between 2e and 2i if the latter interacts with the aromatic ring via C-Cl/π interactions [36]; in this case, the side-chain may adopt a different conformation and affect the dynamics of TMH 6. Despite the differences in the pattern of interactions, the type and frequency of polar and non-polar contacts were fundamentally the same in both ligands, which may explain their similar Ki values. When all the interactions were considered, only two residues of DOR show unique interactions with these ligands: N314 (only with 2e) and S135 (only with 2i). Therefore, substituents that interacted with S135 (or engaged TMH 3 near this residue) and interacted less strongly with N314 (or disengaged TMH 7) may confer potent full DOR agonism. The difference in atomic size, polarizability, and electronegativity, as well as the putative C-Cl/π interactions, appear to play a role in the difference between 2e and 2i. Accordingly, it would be of interest to see the effects of p-Br and p-I on DOR behavior. In MOR, the patterns of interactions of 2i were similar to those in DOR ( Figure 2B): one conformer interacted with TMH 6 and 7 and the other with TMH 3 and 7. This may be enough to impart potent agonism, as in the DOR, but only partial agonist activity in MOR. When all the interactions were considered, F289 is the only residue that was unique in the interaction patterns of 2i with the receptors. The present investigation indicated that changing the balance of interactions of the phenethyl ring with S154, N332, and F289 by a suitable substituent may help design potent, full DOR agonists.
The constrained environment of the tail substituent on the aromatic ring of the N-phenethyl moiety located deep inside the receptor pocket suggested that these interactions can induce major changes in the OR structure or dynamics. The consequences of such effects were already observed in a previous study where a p-NO 2 substitution in the ring elicited significant changes in OR activity and efficacy [8]; computational studies of other GPCRs have also shown the importance of the lower strata of the binding pocket to affect function [37]. Thus, moieties in specific positions on the phenyl ring in N-phenethylnorhydromorphone might convert a potent MOR agonist to MOR antagonist or significantly change its potency. This was shown using in vitro assays with the nitro substituent on the phenyl ring (in 2a and 2b, Table 1). Depending on the position of the nitro substituent in the phenyl ring, one of the compounds (with a p-nitro substituent, 2a) was a potent low efficacy MOR agonist with subnanomolar affinity and the other (with a m-nitro substituent, 2b) was a high-affinity MOR antagonist in the [ 35 S]GTPγS assay (Table 1), a minor positional change inducing a significant change in activity. In the forskolin-induced cAMP accumulation assay, a major difference in potency was observed with these compounds.

Respiratory Depression Assays in Mice
The p-chlorophenyl compound 2i was among the most interesting of the compounds in that it exhibited high MOR and DOR affinity and potency and was a potent efficacious DOR agonist and a partial MOR agonist in the [ 35 S]GTPγS assay (Table 1). It was an exceptionally potent MOR agonist in the forskolin-induced cAMP accumulation assay ( Table 2). Few compounds have been noted in the literature that combine potent MOR partial agonist and potent DOR full agonist activity in a δ/µ ratio of about 7 in opioid receptor binding studies and in a δ/µ ratio of about 1 from potency studies in the [ 35 S]GTPγS assay (Table 1). We thought that it would be of interest to further examine that compound in vivo.
In mice, 2i did not depress respiration rate in the presence of normal air. Figure 3A shows time courses of saline, morphine (10 mg/kg), and different doses of 2i on respiration rate. Figure 3B shows the calculated area under the curve (AUCs) from 6 min to 45 min post injection. As seen in Figure 3B, 10 mg/kg morphine significantly reduced (p < 0.0001) respiration rate compared to saline (One-way ANOVA revealed a significant effect for treatment F(5,38) = 18.34, p < 0.0001).

compound in vivo.
In mice, 2i did not depress respiration rate in the presence of normal air. Figure 3A shows time courses of saline, morphine (10 mg/kg), and different doses of 2i on respiration rate. Figure 3B shows the calculated area under the curve (AUCs) from 6 min to 45 min post injection. As seen in Figure 3B, 10 mg/kg morphine significantly reduced (p < 0.0001) respiration rate compared to saline (One-way ANOVA revealed a significant effect for treatment F(5,38) = 18.34, p < 0.0001). A B Figure 3. Effects of morphine and 2i on respiratory rate in mice. After acclimation in observation boxes, mice were injected with either saline, morphine 10 mg/kg, or 2i and connected to a throat sensor. Five min later, the recording was started and respiratory rate was measured from 6 min to 45 min post-injection (A). Area under the curve (AUC) was calculated from 6 min to 45 min. Morphine significantly reduced respiratory rate compared to saline (B). Data are expressed as mean ± standard error of the mean (SEM.) (n = 6-8) (**** p <0.0001). One-way ANOVA followed by Dunnett's multiple comparison test.
The doses chosen were based on the squirrel monkey tail withdrawal latency assay and the highest dose (0.1 mg/kg) was about 5 or 6 times higher than the ED50 values at 50 and 52 °C from the tail withdrawal latency assay (the usual dose studied to observe side-effects is about 4× the ED50). Compound 2i (0.01-0.1 mg/kg) had no effect on respiration rate in this assay in mice although Figure 3. Effects of morphine and 2i on respiratory rate in mice. After acclimation in observation boxes, mice were injected with either saline, morphine 10 mg/kg, or 2i and connected to a throat sensor. Five min later, the recording was started and respiratory rate was measured from 6 min to 45 min post-injection (A). Area under the curve (AUC) was calculated from 6 min to 45 min. Morphine significantly reduced respiratory rate compared to saline (B). Data are expressed as mean ± standard error of the mean (SEM.) (n = 6-8) (**** p < 0.0001). One-way ANOVA followed by Dunnett's multiple comparison test.
The doses chosen were based on the squirrel monkey tail withdrawal latency assay and the highest dose (0.1 mg/kg) was about 5 or 6 times higher than the ED 50 values at 50 and 52 • C from the tail withdrawal latency assay (the usual dose studied to observe side-effects is about 4× the ED50). Compound 2i (0.01-0.1 mg/kg) had no effect on respiration rate in this assay in mice although morphine, as expected, significantly decreased respiratory rate. Results for oxygen saturation (SpO 2 ) indicated that neither morphine nor 2i had any effect on SpO 2. from 6 min to 45 min post-injection (data not shown).

Antinociceptive Studies and Respiratory Depression Studies in Squirrel Monkeys
Further studies evaluated the effects of 2i and, for comparison, morphine, in assays of antinociception and respiratory depression in nonhuman primates. In a squirrel monkey tail withdrawal latency assay, the p-chlorophenethyl compound 2i exhibited the observed partial MOR agonist in vitro characteristics (in the [ 35 S]GTPγS assay) by having a full effect at moderate (50 • C) and hot (52 • C) but not at very hot (55 • C) water temperatures. In comparison, morphine elicited full antinociceptive effects at all three water temperatures. The compound 2i, like morphine, also produced dose-related behavioral impairment, evident as decreases in operant performance and, consequently, the number of reinforcers obtained during an operant task interspersed between tail-withdrawal tests. The doses of 2i that produced behavioral impairment were similar to those that produce antinociception in 50 • C and 52 • C water (Figure 4).
Molecules 2020, 25, x 18 of 33 morphine, as expected, significantly decreased respiratory rate. Results for oxygen saturation (SpO2) indicated that neither morphine nor 2i had any effect on SpO2. from 6 min to 45 min post-injection (data not shown).

Antinociceptive Studies and Respiratory Depression Studies in Squirrel Monkeys
Further studies evaluated the effects of 2i and, for comparison, morphine, in assays of antinociception and respiratory depression in nonhuman primates. In a squirrel monkey tail withdrawal latency assay, the p-chlorophenethyl compound 2i exhibited the observed partial MOR agonist in vitro characteristics (in the [ 35 S]GTPγS assay) by having a full effect at moderate (50 °C) and hot (52 °C) but not at very hot (55 °C) water temperatures. In comparison, morphine elicited full antinociceptive effects at all three water temperatures. The compound 2i, like morphine, also produced dose-related behavioral impairment, evident as decreases in operant performance and, consequently, the number of reinforcers obtained during an operant task interspersed between tailwithdrawal tests. The doses of 2i that produced behavioral impairment were similar to those that produce antinociception in 50 °C and 52 °C water (Figure 4). The ED50 doses for producing antinociception at 50 °C or 52 °C were, respectively, 0.010 and 0.015 mg/kg, and the ED50 dose for reducing the number of reinforcers earned (behavioral impairment), was 0.022 mg/kg, resulting in ED50 ratios (behavioral impairment/ antinociception) that were about 2 or less. Ratio values greater than 1 imply some behavioral selectivity in observed effects although, as can be seen in Table 3, morphine had an even greater behavioral impairment/antinociception potency ratio than did 2i. Further studies evaluated whether 2i differed from morphine in its capacity to produce respiratory depression in squirrel monkeys. The ED 50 doses for producing antinociception at 50 • C or 52 • C were, respectively, 0.010 and 0.015 mg/kg, and the ED50 dose for reducing the number of reinforcers earned (behavioral impairment), was 0.022 mg/kg, resulting in ED 50 ratios (behavioral impairment/antinociception) that were about 2 or less. Ratio values greater than 1 imply some behavioral selectivity in observed effects although, as can be seen in Table 3, morphine had an even greater behavioral impairment/antinociception potency ratio than did 2i. Further studies evaluated whether 2i differed from morphine in its capacity to produce respiratory depression in squirrel monkeys.
The results from a respiratory depression assay in squirrel monkeys correlated somewhat with data obtained using mice. As shown in Figure 5 during exposure to air alone, neither 2i (0.003-0.1 mg/kg) nor morphine (0.03-3.0 mg/kg) had effects on respiratory rate or overall ventilation (minute volume) in squirrel monkeys, whereas in mice morphine (10 mg/kg), but not 2i (0.01-0.1 mg/kg), depressed ventilation. In contrast, during exposure to 5% CO 2 mixed in air (hypercapnia), both 2i and morphine significantly decreased ventilation in squirrel monkeys, resulting from respiratory depressant effects ( Figure 5). The results from a respiratory depression assay in squirrel monkeys correlated somewhat with data obtained using mice. As shown in Figure 5 during exposure to air alone, neither 2i (0.003-0.1 mg/kg) nor morphine (0.03-3.0 mg/kg) had effects on respiratory rate or overall ventilation (minute volume) in squirrel monkeys, whereas in mice morphine (10 mg/kg), but not 2i (0.01-0.1 mg/kg), depressed ventilation. In contrast, during exposure to 5% CO2 mixed in air (hypercapnia), both 2i and morphine significantly decreased ventilation in squirrel monkeys, resulting from respiratory depressant effects ( Figure 5). The ED50 values for producing antinociception in 52 °C water, decreasing the number of reinforcers earned in an operant behavioral task, and decreasing ventilation in 5% CO2, as well as calculated potency ratios across the procedures, are summarized in Table 3. Here, a complicated picture emerges in which morphine has a larger, and hence more favorable, potency ratio for  Figure 5. Effects in squirrel monkeys of 2i and morphine on respiratory rate (bottom panels) or on minute volume (top panels) in the presence of normal air (open symbols) or air mixed with 5% CO 2 (filled symbols). 2i and morphine significantly reduced minute volume in 5% CO 2 without significantly altering respiratory rate. Data are expressed as mean ± SEM. (n = 4); (*) indicates the difference from saline (p ≤ 0.01; one-way ANOVA followed by Dunnett's multiple comparison test. The ED 50 values for producing antinociception in 52 • C water, decreasing the number of reinforcers earned in an operant behavioral task, and decreasing ventilation in 5% CO 2 , as well as calculated potency ratios across the procedures, are summarized in Table 3. Here, a complicated picture emerges in which morphine has a larger, and hence more favorable, potency ratio for antinociceptive and behaviorally disruptive effect than does 2i, whereas 2i has a higher ratio for behaviorally disruptive and respiratory depressant effects. Indeed, the potency ratio for morphine for behavioral disruption and respiratory depression was ≤ 1, indicating that breathing was decreased at similar doses to those that decreased behavior, whereas 2-fold higher doses of 2i were needed to decrease CO 2 -stimulated breathing ( Figure 5).
The ED 50 values for producing antinociception in 52 • C water, decreasing the number of reinforcers earned in an operant behavioral task, and decreasing ventilation in 5% CO 2 , as well as calculated potency ratios across the procedures, are summarized in Table 3. Here, a complicated picture emerges in which morphine has a larger, and hence more favorable, potency ratio for antinociceptive and behaviorally disruptive effect than does 2i, whereas 2i has a higher ratio for behaviorally disruptive and respiratory depressant effects. Indeed, the potency ratio for morphine for behavioral disruption and respiratory depression was ≤1, indicating that breathing was decreased at similar doses to those that decreased behavior, whereas 2-fold higher doses of 2i were needed to decrease CO 2 -stimulated breathing.

Assays
These were performed as previously described using HitHunter CHO-K1 cells expressing either human OPRM1, OPRK1, or OPRMD1 cells. Briefly, cells were dissociated from culture plates and plated at 10,000 cells/well in a 384-well tissue culture plate and incubated overnight at 37 • C in 5% CO 2 . Stock solutions of compound were made in 100% DMSO at a 5 mM concentration. A serial dilution of 10 concentrations was made using 100% DMSO, creating 100× solutions of the compound for treatment. The 100× solutions were then diluted to 5× solutions using assay buffer consisting of Hank's Buffered Salt Solution, HEPES, and forskolin. The HitHunter cAMP Assay for Small Molecules by DiscoverX was then used according to manufacturer's directions, utilizing the 5× solutions containing the compound studied. Cells were incubated with compound for 30 min at a 1× concentration. The following day, the Cytation 5 plate reader and Gen5 Software were used to quantify luminescence (BioTek, Winooski, VT, USA) [41].

β-Arrestin-2 EFC Recruitment Assay
Assays were performed as previously described [41] using PathHunter human µ-opioid receptor β-arrestin-2 EFC cells. Briefly, cells were dissociated from culture plates and plated at 5000 cells/well in a 384-well tissue culture plate and incubated overnight at 37 • C in 5% CO 2 . Stock solutions of compound were made in 100% DMSO for a final concentration of 5 mM. A serial dilution of 11 concentrations was made using 100% DMSO to create 100× solutions of the compound. Assay buffer containing Hank's Buffered Salt Solution and HEPES was used to dilute the 100× solutions to 5× solutions. The DiscoverX PathHunter assay was used according to manufacturer's instructions. Cells were treated with the compounds for a final 1× concentration for 90 min. at 37 • C and 5% CO 2 . Reagents from the assay kit were used accordingly and the cell culture plate was protected from light for 1 h. Cytation 5 plate reader and Gen5 Software were used to quantify luminescence (BioTek).

Data Analysis
Data were analyzed as previously described [42] using GraphPad Prism 6.0 software (GraphPad, San Diego, CA, USA different address given before). Briefly, sigmoidal dose-response curves in the forskolin-induced cAMP accumulation assay and the β-arrestin2 EFC recruitment assay were generated using nonlinear regression analysis. Compounds were evaluated in triplicate in individual experiments with n ≥ 2. All values in the cAMP accumulation assay and β-arrestin2 recruitment assay are reported as the mean ± SEM. Bias factors were calculated using Equation (1) shown below.
Log (bias f actor) = Log EC50 test×Emax DAMGO air or a 5% CO 2 in air mixture) was introduced to and extracted from the chamber at a constant flow rate of 5 L/min. Experimental sessions consisted of 4-6 consecutive 30 min cycles, each comprising a 15 min exposure to air followed by a 15 min exposure to 5% CO 2 . Drug effects were determined using cumulative dosing procedures, and injections were administered following each exposure to 5% CO 2. Respiratory rate and tidal volume (mL/breath) were recorded over 1 min periods and were multiplied to provide minute volumes. Data from the last three minutes of each exposure to air or CO2 were averaged and used for analysis of drug effects on ventilation.

Computational Methods
All the ligands considered in this study and their conformers (see Table S1 and Figure S2A in SI) in their protonated form were geometry optimized via quantum chemical (QM) calculations at the B3LYP/6-31G* level in the gaseous phase as implemented in Gaussian 09 software [45]. The atomic polar tensor derived charges from these calculations were used to assign a partial charge on each atom for the ligands. All other parameters were determined by chemical analogy with the topology and the parameters files of the all-atom CHARMM force field (version c42b2) [46]. The structures of the inactive forms of MOR (PDB # 4DKL) and DOR (4EJ4) were taken as the starting configurations. After removing the cocrystallized ligands and crystal water/ions, the intracellular loop (ICL3) connecting TM5 and TM6 was first modeled in each receptor using MM (ab initio) methods [8,47,48]. The modeled receptors were then embedded in a membrane composed of zwitterionic 1-palmitoyl-2-oleoyl-phosphatidylcholine lipid molecules with initial concentrations of Na + , K + , and Clions in the extracellular (EC) and intracellular (IC) regions. All the amino acids were assumed to be neutral at physiological pH except Asp -, Glu -, Lys + and Arg + ; additional ions were used to neutralize the systems. The initial orientation and relative position of the receptor with respect to the membrane were obtained from the OPM database [49]. The receptor and the membrane were then solvated in TIP3P water, and a single Na + ion was introduced in the binding pocket and coordinated as in the DOR (4N6H) [50]. The system was then energy minimized and thermally equilibrated according to the following protocol: first, the conformation of the receptor and Na + were kept fixed, and the membrane and water were gradually heated to the target temperature (37 • C) at constant pressure (1 atm); the system was then equilibrated for 5 ns; the constraints on the side chains and ion were then removed, and the system equilibrated for another 5 ns; finally, all the constraints were removed, and the system equilibrated for another 5 ns. Throughout the entire equilibration process, the ion remained coordinated with the anchoring aspartic acid D147 (MOR) and D128 (DOR) (sequence numbering follows the corresponding crystal structures). A conformation (snapshot) of each system at the end of the final equilibration phase was used to dock the ligands. These are the basal conformations that all the ligands "see" before entering the binding pocket, and were used for all the comparative analyses, regardless of the experimentally determined activity or pharmacological outcome. After removing the Na + ion, the ligands (and their conformers) were rigidly docked into the binding pocket based on two criteria: the close contact between the charged amine and the anchoring Aspand the binding mode of the antagonists β-FNA and naltrindole co-crystallized with the MOR and the DOR, respectively. Several conformers (cf. Figure S2) could be eliminated by steric considerations alone. Others (especially those involving rotations of φ 1 ; cf. Figure S2) could still dock without apparent steric clashes upon small relaxations of the pocket side chains, but they tended to lose the critical interaction with Aspin the course of equilibration or early production; when this occurred in repeated simulations, the conformer was discarded. Overall, between one and four stable conformers were left for each ligand (see Table S2). The three-stage equilibration protocol described above for Na + was repeated for each ligand/conformer after docking. Steric relaxation of the ligand and residues in the pocket set in during the early stages of equilibration. Five independent 50 ns molecular dynamics simulations were conducted for each system at 37 • C and 1 atm, using periodic boundary conditions and particle mesh Ewald summations. This simulation time was sufficient to ensure convergence and statistical analysis of the quantities of interest (hydrophobic contacts, H-bonds, electrostatic interactions), which were computed after structural relaxation set in (estimated from Cα-RMSD vs. time), typically during the last half of the dynamic trajectory (production run). The results combine data from all the independent simulations for each ligand conformer/OR.

Conclusions
We hypothesized that substituents at the tail end of the body of a large molecule might modify the in vitro activity and/or potency of the compound and possibly modify a G-protein biased compound that acted primarily through MOR to a bifunctional ligand. We found that a substituent, a chlorine atom, modified the activity of N-phenethylnorhydromorphone (S11), a potent full agonist with a DOR-MOR δ/µ potency ratio of 38.5, to a compound with a δ/µ potency ratio of 1.2, N-p-chloro phenethylnorhydromorphone (2i). It exhibited potent partial MOR agonist and potent full DOR agonist activity. In fact, the introduction of a p-Cl substituent (2i) in the N-phenethyl moiety did not particularly reduce the MOR potency of S11 but instead increased its DOR potency; it induced a change from a molecule that acted primarily as a MOR ligand to a bifunctional compound with the ability to interact potently with MOR and DOR. This change was due to a simple substituent at the tail end of the compound. Molecular modeling and simulations found that the substituent on the aromatic ring of the N-phenethyl moiety is located in an area where relatively small changes in the N-phenylethyl ring via substitution perturb residues located in quite different regions of the opioid receptors and engage different TMHs. In theory, the combination of MOR and DOR properties found in 2i might have made the compound less likely than other potent analgesics to cause respiratory depression [17]. Indeed, that was found to be the case in mice using normal air, where a clear difference was found between the effects of 2i and morphine on respiration. Both 2i and morphine are partial agonists; if the lack of effect on respiration was due to the partial agonist character of 2i, the same effect would be expected with morphine. The ability of 2i to recruit β-arrestin2 at least as well as morphine ( Table 2) would predict that it should exhibit all of the side-effects known to occur with morphine. The inability of 2i to depress respiration in mice might indicate that the recruitment of β-arrestin2 may not be the cause of all of the side-effects seen with opioids [4]. However, with squirrel monkeys under more stringent conditions, in an assay using 5% CO 2 mixed in air, 2i was found to be as effective as morphine in depressing respiration. Further work is necessary to determine whether 2i will produce the gastrointestinal effects, tolerance, and dependence that occur with other G-protein biased opioids.   13 C-NMR spectra of novel compounds, molecular modeling, and dynamics simulations, and Scheme S1, the Synthesis of hydromorphone (S5). N-norhydromorphone (S10) and N-phenethy lnorhydromorphone (S11). Funding: The work of SI: EBG, ad MWA was supported by NIDA grant P30 DA13429. The work of JRT was supported by NIDA grant DA039997. The work of SK, RSC and TEP was supported by NIDA grant DA018151. The work of SLW, CAP, and JB was supported by NIDA grants DA035857 and DA047574.The work of MW, TCI, CAH, AEJ and KCR was supported by the NIH Intramural Research Programs of the National Institute on Drug Abuse and the National Institute of Alcohol Abuse and Alcoholism. The work of TAK and JLK was supported by the NIH Intramural Research Programs of the National Institute on Drug Abuse. The work of SAH and YSL was supported by the NIH Intramural Research Program through the Center for Information Technology. The computational studies utilized PC/LINUX clusters at the Center for Molecular Modeling of the NIH (http://cmm.cit.nih.gov) and the computational resources of the NIH HPC Biowulf cluster (http://hpc.nih.gov). The work of TAK and JLK was supported by the NIH Intramural Research Programs of the National Institute on Drug Abuse. We thank the NIDA Drug Supply for providing compounds used in the forskolin-induced cAMP assays. We also thank John Lloyd and Noel Whittaker (Mass Spectrometry Facility, NIDDK) for the mass spectral data, and S. Steven Negus (Dept. of Pharmacology and Toxicology. Virginia Commonwealth University, Richmond VA) for helpful discussions about the possible pharmacological effects of compounds with MOR and DOR agonist activity.

Conflicts of Interest:
The authors declare no conflict of interest.