Optimization of a Pyrimidinone Series for Selective Inhibition of Ca2+/Calmodulin-Stimulated Adenylyl Cyclase 1 Activity for the Treatment of Chronic Pain

Adenylyl cyclase type 1 (AC1) is involved in signaling for chronic pain sensitization in the central nervous system and is an emerging target for the treatment of chronic pain. AC1 and a closely related isoform AC8 are also implicated to have roles in learning and memory signaling processes. Our team has carried out cellular screening for inhibitors of AC1 yielding a pyrazolyl-pyrimidinone scaffold with low micromolar potency against AC1 and selectivity versus AC8. Structure-activity relationship (SAR) studies led to analogues with cellular IC50 values as low as 0.25 μM, selectivity versus AC8 and other AC isoforms as well as other common neurological targets. A representative analogue displayed modest antiallodynic effects in a mouse model of inflammatory pain. This series represents the most potent and selective inhibitors of Ca2+/calmodulin-stimulated AC1 activity to date with improved drug-like physicochemical properties making them potential lead compounds for the treatment of inflammatory pain.


INTRODUCTION
Adenylyl cyclases (ACs) are effector enzymes downstream of various G protein-coupled receptors (GPCRs) and ion channels that transduce signals via the catalysis of adenosine triphosphate (ATP) to cyclic adenosine 3',5'-monophosphate (cAMP). 1 In line with the many roles of the numerous GPCRs and ion channels, modulation of ACs leads to a variety of physiological effects dependent on the AC isoform, interaction partners, and tissue localization. Humans encode nine membrane-bound ACs that are organized into four groups based on their regulatory mechanisms to various intracellular stimuli. 2,3 Group 1 ACs include AC1, AC3, and AC8 and are characterized by their positive modulation by Ca 2+ /Calmodulin (CaM) although AC3 is conditionally stimulated by this complex and requires the presence of additional G protein subunits. 4 Group 2 ACs include AC2, AC4, and AC7 and are conditionally activated by G protein βγ (Gβγ) subunits. 5 Group 3 ACs include AC5 and AC6 and are negatively modulated by Ca 2+ . 6 Finally, Group 4 contains only AC9, which is the lone isoform relatively insensitive to forskolin, an allosteric agonist of ACs.
Processes such as memory acquisition, drug tolerance and dependence, and chronic pain are known to be impacted significantly by ACs. 7,8 Although ACs are expressed ubiquitously throughout the body, certain isoforms have distinct tissue expression patterns. 9 AC1 and AC8 are primarily expressed in the central nervous system (CNS) within regions such as, but not limited to, the hippocampus and the anterior cingulate cortex (ACC); regions of the brain associated with learning, memory and the development of chronic pain. [9][10][11][12] Evidence suggests AC1 is responsible for propagation of inflammatory pain stimuli. 13 In vivo studies indicate that injury leads to increased postsynaptic Ca 2+ influx in the ACC, where Ca 2+ forms a complex with CaM that, in turn, binds and subsequently activates AC1 to produce cAMP. 14,15 In chronic pain, it is suggested that persistent Ca 2+ influx causes hyperactivation of AC1 and downstream pain sensitization. 16 Moreover, AC1 knockout (AC1 -/-) and AC1/8 double knockout (DKO) mice exhibit nearly complete abrogation of behavioral pain response when treated with an inflammatory cocktail, complete Freund's adjuvant (CFA), and display a lack of pain sensitization in a muscle pain model. 13,17,18 However, both AC1 and AC8 are implicated to play a role in long-term memory and long-term potentiation in these brain regions. [19][20][21] In mouse knockout models, DKO mice displayed severe impairment to spatial memory acquisition; however, this impairment was mostly absent in the AC1 -/mice. 19,[22][23][24][25] Additionally, ACs propagate signals downstream from the µ-opioid receptor (MOR), a well characterized target for analgesic pharmacological therapy. MOR agonists act, in part, by negatively regulating AC1. 18 Upon agonist stimulation the MOR activates and induces dissociation of a heterotrimeric G protein complex comprised of G protein αi (Gαi) and Gβγ subunits. The Gαi subunit can then translocate to a membrane bound AC, in this case AC1, and subsequently inhibit catalysis and reduce intracellular cAMP levels. 26 Additionally, chronic opioid administration leads to compensatory neuroadaptation, including upregulation of AC1 to offset the reduced cAMP signal and ultimately leading to tolerance and dependence. 18,27,28 Moreover, activation of the MOR signaling pathway has several undesired side effects, some of which are mediated by recruitment of β-arrestins and the release of Gβγ, which can lead to the development of physiological tolerance and may also contribute to opioid-induced respiratory depression, respectively. [29][30][31] Furthermore, MOR expression is not limited to the brain but is also expressed peripherally, leading to other common side effects such as opioid-induced constipation. 32 One potential approach under investigation to limit these side effects is the development of biased MOR agonists that favor signaling through the G protein pathway and reduce the β-arrestin pathway signal. 33 Alternatively, our group and others have shown it is possible to target the activity of the downstream AC directly, bypassing the MOR entirely. [34][35][36] This strategy would allow for effective inhibition of chronic pain through AC inhibition, while avoiding negative side-effects induced by clinically used MOR agonists; thus, increasing the therapeutic index. Taken together, the promising genetic evaluation coupled with the tissue localization at the key pain center of the brain and its role in analgesic response to opioid agonists all suggest AC1 is a promising target for development of novel pharmacological modulators for the treatment of chronic pain.
Based on the aforementioned in vivo genetic data, a key requirement for a suitable AC1 modulator for chronic pain therapy is that it must be selective for AC1 versus AC8 to avoid the potential for memory impairment. Our group and others have since screened and tested for such modulators; the resultant known AC1 inhibitors and modulators include the adenosine-based SQ22536, 37 the adenosine-based NB001, 36 the chromone ST034307, 34 and the oxadiazole AC10065 35 (Figure 1). However, each molecule presents drawbacks and challenges for potential use as a pharmacotherapeutic agent. SQ22536 is proposed to bind the catalytic site of AC, and while it displays cellular potency of >10 µM (IC50) versus AC1 it lacks selectivity versus other isoforms including AC5. 38 NB001 likewise has only 14-fold selectivity for AC1 versus AC8 in cell-based assays. Given that both of these compounds have adenine moieties, there is concern for selectivity versus other adenine-binding and ATP-binding proteins that could lead to off-target effects in vivo. 39 Our group previously published the chromone ST034307, which has a 2.3 µM IC50 versus Ca 2+ /CaM-stimulated AC1 activity in cells with no inhibition of AC8. 34 This selective modulator was then shown to produce analgesic effects versus inflammatory pain in vivo. Despite this activity, the molecule presented significant physicochemical liabilities, namely poor aqueous solubility which hampered dosing. Furthermore, SAR was found to be intractable beyond the published data. In the interest of finding a scaffold with tractable SAR and better drug-like characteristics, a high-throughput screen was undertaken followed by SAR elucidation of the oxadiazole series, yielding the most recent AC1 modulator AC10065. 35 This compound inhibited AC1-mediated Ca 2+ /CaM-stimulated cAMP production in cells with an IC50 of 1.4 µM versus AC1 and 4.1 µM versus AC8, and displayed modest in vivo efficacy in a mouse CFA inflammatory pain model. 35 However, the oxadiazole series was once again limited in dosing due to the scaffold's poor aqueous solubility. In summation, these current AC1 modulators suggest that AC1 inhibition can produce anti-allodynic effects in behavioral animal studies and that isoform selectivity between AC1 and AC8 is achievable.
Building upon our previous work, we identified a pyrimidinone scaffold of AC1 modulators from the same high-throughput screen used to discover the oxadiazole series. These pyrimidinone AC1 modulators, represented by hit 1, displayed similar potency as ST034307 and AC10065 while maintaining selectivity versus AC8. After resynthesizing and validating these hits, they were found to be potent and selective for inhibiting Ca 2+ /CaM-stimulated AC1 activity over AC8, with maximal inhibition versus AC8 of 25-50% at the highest concentrations tested. Our team then designed and synthesized the following series of pyrimidinone analogs to elucidate the SAR, improve potency versus AC1, and maintain selectivity. Following activity testing a prioritized analog was chosen for in vivo testing in the behavioral CFA-inflammatory pain model. Our results detailing the discovery of a sub-micromolar inhibitor of AC1-mediated Ca 2+ /CaM cAMP production with selectivity versus AC8 and other AC isoforms are described below.

High-throughput screen for the discovery of inhibitors of Ca 2+ /CaM stimulated activity of AC1
To discover additional potent and selective inhibitors of AC1 activity, our team designed and carried out a high-throughput screen of 10,240 compounds from the Life Chemical diversity library. Our primary screen sought to identify molecules that inhibit AC1 mediated Ca 2+ /CaMstimulated cAMP production in HEK cells stably expressing AC1 (HEK-AC1). HEK-AC1 cells were stimulated with the Ca 2+ ionophore A23187 and the accumulation of cellular cAMP levels was quantified using a homogenous time resolved fluorescence (HTRF) assay with cAMP detection reagents as described previously. 34 Two representative hit compounds per cluster were selected to further validate the inhibitory effects of each structural scaffold on AC1 activity and counter-screened for AC8mediated Ca 2+ /CaM-stimulated cAMP production. Dose-response curves were generated for each compound in HEK-AC1 cells and HEK-AC8 cells to quantify potency and AC8 selectivity. Hits were also assessed for cell viability using CellTiter-Glo to eliminate potential false-positives as a result of cell toxicity. Four clusters appeared to be false positives and were removed from further consideration. An additional cluster was excluded from further analysis because the compounds interfered with the fluorescence emission at 620 nm of the cAMP HTRF detection technology. The compounds from the remaining four scaffolds showed a dose-dependent inhibitory effect on AC1mediated Ca 2+ /CaM-stimulated cAMP production with IC50 values in the low micromolar range.
Distinctively, the confirmed hit compounds of two clusters, an oxadiazole scaffold and a pyrimidinone scaffold, stood out from the rest of hits as their IC50 values were in the single-digit micromolar range and the compounds displayed no apparent toxic effects on the HEK-AC1 cells after 2-hour incubation. Results from optimization of the oxadiazole scaffold have since been reported leading to the development of AC10065. 35 The other single-digit micromolar potency scaffold was the pyrimidinone series represented by hit 1 (Figure 1) and the concentrationresponse curves for against AC1 and AC8 are presented in Figure 2.

Chemistry
The general structure of the pyrimidinone hit cluster was a 3-ring aromatic scaffold comprised of a pyrimidinone, pyrazole, and an amide-coupled phenyl for which a 5-step linear synthetic sequence was designed. The pyrimidinone portion of the scaffold (Scheme 1) was synthesized via initial cyclocondensation of thiourea and ethyl 3-oxopentanoate to produce 6-ethyl thiouracil intermediate (2). 42 Intermediate 2 was selectively S-methylated with a slight excess of methyl iodide starting at 3 °C and was allowed to warm to room temperature producing intermediate 3. 43  able furnish low to moderate yields for the final amide products 1, 6, 9-32, 36-40. The methoxymethyl containing analog 32 was deprotected to yield the phenol 33. Yields for analogs with electron-rich benzoic acids (e.g. 21-26) were below the quantity necessary for biological testing, thus further optimization led to a final reaction using lithium bis(trimethylsilyl)amide (LiHMDS) pre-activation of the primary amine alongside in situ benzoyl fluoride activation of benzoic acids to provide yields sufficient for testing for these more synthetically-challenging analogs.
Several analogs required alternate methods to prepare (Scheme 2). Compound 7 was a product of an unintended side reaction with 2-acetoxybenzoic acid, which yielded only the acetyl substituted arylamine rather than the intended benzamide coupling product. Compound 8 was prepared via traditional HATU amide coupling as the acyl fluoride procedure did not yield the intended product. Finally, analogs 34 and 35 were found to be incompatible with either acyl fluoride coupling procedure and no desired product was obtained. In the case of 35, according to the literature 46 carboxylic acids with α-protons were found to be incompatible with the acyl fluoride method due to competing ketene generation (despite successful synthesis of later analog 36). In the case of 34, the electron-donating character of the 4-dimethylamino substituent on the carboxylic was likely a contributing factor to the inability to isolate product. This led us to scout several amide coupling methods and ultimately the use of a mixed anhydride coupling method using propylphosphonic anhydride (T3P) 47

Structure-Activity Relationship Studies
All analogs were tested for inhibition of both AC1-and AC8-mediated Ca 2+ /CaMstimulated cAMP production in a cellular assay. For these activity assays, HEK293 cells had endogenous AC3 and AC6 isoforms knocked out using CRISPR-Cas9 followed by subsequent stable transfection and overexpression of AC1 or AC8 isoforms. 48 SAR was informed based on potency against AC1-mediated cAMP production. However, in terms of evaluating selectivity against AC8 a simple comparison of IC50 values does not adequately convey this measure. This is because in the concentration-response curves it was observed that inhibition often reaches baseline levels for AC1 at the higher molecule concentrations while AC8 is never fully inhibited by any analog (representative curves for 1 shown in Figure 2). This indicates that the scaffold displays a difference in maximal inhibitory efficacy between isoforms; therefore, merely evaluating the AC8 IC50 to assess selectivity is misleading as the IC50 is generated from the relative maximum and minimum cAMP signal for AC8. Case in point, analog 1 has a calculated IC50 value of 1.6 µM against AC8 even though it clearly is never fully inhibited. Thus, a more accurate measure of selectivity for AC1 over AC8 would be the evaluation of the percent inhibition of AC8 at the IC90 of AC1 activity (Figure 2, black arrow). Therefore, the AC8 activity will be referred to in this context. All AC1 IC90s are reported in Table S1 and concentration response curves for all molecules against AC1 and AC8 are included in the supporting information.
Initial hit compound 1 was identified in the primary screen as a potent, selective inhibitor of Ca +2 /CaM-stimulated AC1 activity. After re-synthesis and validation, it was found to have a cellular IC50 of 1.4 µM versus AC1-mediated cAMP production and 46% inhibition versus AC8mediated cAMP production at the AC1 IC90. This AC1 potency was comparable to, or an improvement upon, the prior art for AC1 inhibitors and this scaffold was prioritized for optimization. Removal of the fluorine to yield unsubstituted phenyl derivative 6 displayed slightly reduced potency with an IC50 of 2.4 µM versus AC1. We then evaluated a group of analogs containing phenyl ring replacements in compounds 7-9. The phenyl ring was removed leaving an acetamide in compound 7, which showed a complete loss of activity against both ACs. Two bioisostere substitutions, a furan (8) and a pyridine (9), were assessed next. The furanyl substitution 8 was tolerated, albeit with reduced potency (AC1 IC50 value of 6.7 µM), whereas the pyridinyl analog 9 displayed similar activity to the phenyl (AC1 IC50 value of 1.2 µM) indicating the pyridine may be a suitable replacement for the phenyl without detriment to AC1 activity.
However, this molecule did exhibit an increase in AC8 activity, as compared to the unsubstituted phenyl analog (6) but was comparable to the hit 1. Taken together these analogs suggest optimal potency versus AC1 requires at least a 6-membered hydrophobic ring. Based on these data, we proceeded to expand the SAR exploring substitutions on the phenyl for modulation of AC1 potency and selectivity.
In the first cohort we examined the effects of fluorine, chlorine, and bromine on compound activity. No change in activity was observed moving from 2-F in 1 to 3-F in analog 10. However, moving the fluorine to the para-position (11) provided roughly 2-fold improvement in AC1 potency an IC50 value of 0.79 µM while the AC8 activity was relatively unchanged. Substitution for chlorine at either the meta- (12) or para-position (13) yielded modest 2-to 3-fold improvement of potency compared to the fluorine containing counterparts with AC1 IC50 values of 0.47 and 0.37 µM, respectively. No improvement in selectivity versus AC8 was observed as the analogs still remained around 50% inhibition at the AC1 IC90 value. The 3-Br analog (14) displayed slightly reduced potency toward AC1 (IC50 value of 0.61 µM) compared to the corresponding 3-Cl, however, this analog showed the first boost in selectivity over AC8 with only 15% inhibition of the isoform. Attempts were made to synthesize analogs with chlorine or bromine at the ortho position, however yields were very low for these reactions and did not provide enough material for testing. a Indicated groups were benzene ring replacements. b AC1 IC50s calculated from concentration-response curves with inhibitor versus 3 µM A23187-stimulated cAMP accumulation in HEK AC1 3/6 KO cell lines (n=3+). c AC1 IC90s were calculated from AC1 concentration-response curves and then interpolated through AC8 concentration-response curves (n=2+) and subtracted from 100%. All tabular AC1 IC90 data is provided in Table S1. ND = not determined by Prism software due to wide variance or not enough data points at higher concentrations.
Next, we evaluated SAR on the phenyl ring with lipophilic alkyl substituents. Similar to the halogen series, the ortho substituted analogs were less potent against AC1 compared to the meta and para counterparts as illustrated by a 5-fold improvement in AC1 activity when moving from the 2-methyl (15, IC50 = 2.2 µM) to the 3-methyl (16, IC50 = 0.41 µM, Table 1 and Figure   3A). The improvement was more pronounced with the larger ethyl series as the 3-ethyl (20)  inhibition) compared to the halogenated analogs (15 -55%). Interestingly, swapping the 3-methyl for 3-trifluoromethyl (18) resulted in a slight reduction of AC1 activity but also reduced the selectivity over AC8 to 41%, which was a value closer to those observed for the halogenated analogs. Increasing the alkyl branching in this series provided mixed results. For example, moving from methyl (16) to ethyl (20) to iso-propyl (22) Figure 3C and D, all remaining dose-response curves are provided in supporting information.
To summarize, the SAR for AC1 potency and AC8 selectivity favored small, alkyl substituents over halogenated analogs for both activity metrics. Increased alkyl branching generally led to both reduced activity against AC1 and selectivity over AC8. Polar functional groups were tolerated but displayed reduced AC1 potency with the exception of the 4-hydroxy, which was the only substituted phenyl analog to display a complete loss of AC1 activity.
Combined modifications produced mixed results with mid-nanomolar range AC1 potencies but slightly improved AC8 selectivity. Several analogs are within a range for AC1 activity and AC8 potency that would make them reasonable to advance toward in vivo efficacy assays, thus, we prioritized analogs 16, 20, 38 and 39 for further evaluation (dose-response curves provided in toxicity in the HEK cells utilized for the assay. However, to ensure that the molecules did not have undesired cytotoxicity we assessed compounds 16, 20, 38, and 39 in the CellTiter-Glo assay when incubated with HEK293 cells at 30 µM for 1 hour. Across the group of five analogs there was no reduction in cell viability indicating the scaffold is non-toxic to this human cell line (Table S2).

Analog activity in alternative AC stimulating condition and versus other AC isoforms
As selectivity is a critical parameter for AC1 inhibitors to be considered for clinical use, we examined the activity profile of the pyrimidinone scaffold for potency against other AC isoforms ( Figure 4, Table S3). Screening versus representative Group I, II, and III AC isoforms showed partial inhibition of Ca 2+ /CaM-stimulated AC8 activity, weak potentiation of Phorbol 12myristate 13 acetate (PMA)-stimulated AC2 activity, and no apparent effect on forskolinstimulated AC5 activity (Figure 4). This data supports that the scaffold is selective for AC1 against not only the Ca 2+ /CaM-stimulated AC8 but also against AC isoforms that are differentially activated. . Inhibitory activity for five analogs against Group-1 ACs, AC1 (blue) and AC8 (white), as well as group-2 AC2 (green) and Group-3 AC5 (gray). Y-axis shown as percent stimulation of cAMP normalized to 100 % for each isoform. X-axis indicates treatment condition or analog tested. Basal indicates cAMP levels prior to stimulation. Stim indicates cAMP levels after stimulation and treated with DMSO control. All analogs tested at 10 µM (n = 3+ replicates). Error bars indicate standard error of the mean.

Activity versus common neurological off-targets
The original hit (1) (Table S5). Likewise, no antagonist activity was observed in the Gq/Ca 2+ assay using a calcium sensitive reporter; however, there was consistent inhibitory activity observed in the antagonist Tango assay with IC50 values ranging from 0.42 -0.60 µM. The somewhat contrasting results in the Gq/Ca 2+ and -arrestin antagonist assays are perplexing and may reflect a complex pharmacology requiring further study or may be due to compound interference with the Tango assay. The latter is supported by data revealing that compounds 20 and 39 that weakly bind the 5-HT2B receptor (<50% displacement at 10 μM; Table S4), yet inhibit in the Tango assay with submicromolar potency.
Given that four analogs displayed relatively selective profiles from the PDSP the data was not collected in full for analog 38, however the molecule was assessed separately for activity against the MOR, KOR, and DOR as these are relevant to the pain signaling process. The data confirms that analog 38 did not display any significant off-target modulation of the ORs at 10 µM (Table S6) similar to 1, 16, 20, and 39. The data as a whole suggests that this compound series is not inhibiting AC1 activity through agonism of a known Gi/o-linked GPCR.

Activity versus off-target kinases and hERG
The scaffold was further evaluated for off-target liabilities against the human Ether-á-gogo related gene (hERG) potassium channel and a select group of kinases. First, the pyrimidinone moiety bears resemblance to previously reported kinase inhibitors. 50,51 To determine the potential liability that off-target kinase inhibition may render we selected four kinases that are earmarked by pharmaceutical companies as 'sentinel representatives' of kinase families that should be assessed for pre-clinical profiling to reduce safety-related drug attrition. 52 Analogs 20, 38, and 39 were selected to be assessed against this small panel for inhibitory activity at a single dose of 10 µM. We found that analogs 20 and 38 did display inhibition of insulin receptor kinase (IRK) at 57% and 64% inhibition, respectively (Table S7). Conversely, analog 39 only exhibited 4% inhibition against IRK. For the remaining kinases, analog 20 displayed 28% inhibition versus vascular endothelial growth factor receptor-2 (VEGFR2) but did not exhibit any inhibition toward leukocyte C-terminal Src kinase (LCK) or Rho-associated coiled-coil containing kinase (ROCK1).
Analogs 38 and 39 did not inhibit any of the remaining three kinases tested. Therefore, while analogs 20 and 38 displayed appreciable activity against IRK the inhibition does not appear to be intrinsic to the scaffold as 39 was inactive against these targets. Nevertheless, it would be prudent to continue to monitor analog effects on kinases and expand the panel for future studies.
Finally, both analogs were also evaluated for ability to bind to the hERG potassium channel to assess for potential cardiac liability. Analog 20 displayed 0% inhibition of [ 3 H]-astemizole binding at a dose of 10 µM while 38 and 39 exhibited 6% and 11% inhibition at the same dose (Table S6). No molecule was determined to be a significant hERG binder.

In silico and in vitro physicochemical and pharmacokinetic properties of prioritized analogs
This class of molecules did display reduced solubility in aqueous media, thus, we first assessed thermodynamic solubility experimentally for the four prioritized analogs using a method previously described. 53,54 This analysis showed 38 was the most soluble at 33.6 µM in PBS at pH 7.4 and 25 °C, which is approximately 60-fold higher than the cellular IC50 for inhibition of Ca2+/Calmodulin-stimulated AC1 activity. This was followed by 16 (19.5 µM), 20 (10.8 µM) and 39 (7.6 µM) in decreasing order of solubility (Table 2).
Next, as a means to select a candidate for evaluation in an in vivo model for inflammatory pain, cheminformatics analysis of physicochemical and pharmacokinetic properties was carried out using QikProp (Schrödinger, LLC). Molecules were loaded into Maestro and prepared for computational analysis. Using the OPLS3e force field the protonation states at pH 7.4 were determined and energy minimization was performed. Since the presumed site of action for our compounds is in the CNS we have given consideration to molecular properties that could affect blood-brain barrier (BBB) permeability. Several analyses have been published on physicochemical properties that drive BBB permeability and in general the consensus is that molecules with molecular weight < 400 g/mol and LogP ranging from 2 -5 have increased probability for CNS activity. [55][56][57][58] The molecular weights of the four prioritized analogs range from 337 g/mol to 355 g/mol and QPLogPo/w ranged from 2.32 -2.88 (specific values for selected analogs provided in   68 Thus, the effect of small molecules on CYPs should be assessed to de-risk the scaffold for lead optimization. To this end, we evaluated 38 for inhibition at five CYPs that contribute to metabolism of over 50% of marketed drugs. 68,69 The molecule was dosed at a single concentration of 10 µM and evaluated for percent inhibition.
In contrast to the pkCSM predictions it was found that 38 had no significant inhibition of any of the five CYPs tested in this panel (

In vivo efficacy of compound 38 in a mouse model for inflammatory pain
To assess whether 38 could reduce pain sensitization we evaluated the analog in a mouse model for inflammatory pain with CFA as previously published. 34  A B

DISCUSSION
Following our previous reports of selective AC1 inhibitors for chronic pain, we sought to discover a scaffold with improved drug-like characteristics and potency while maintaining selectivity versus AC8. The same high-throughput screen that identified the prior oxadiazole series 35 also yielded the pyrazolyl-pyrimidinone scaffold as potent AC1 inhibitors with selectivity against AC8 and having higher predicted aqueous solubility. The optimization was multifaceted as we sought to improve AC1 potency while maintaining/improving selectivity over AC8.
However, during the design we were also cognizant of modifications that may improve solubility and/or BBB permeability. After SAR optimization of the benzamide moiety, we were able to improve potency by 2 -4-fold across the series of analogs versus Ca 2+ /CaM-stimulated AC1 activity while maintaining/improving selectivity versus AC8. The SAR revealed a 6-membered aromatic group was preferred on the amide as both the phenyl and 2-pyridine functional groups were comparable with low-micromolar potency against AC1. We moved forward with phenyl modifications for two reasons: 1) a greater availability of substituted reagents to better explore SAR and 2) we wanted to limit the number of heteroatoms in an attempt to improve BBB permeability. Hydrophobic substituents were preferred on the phenyl ring, particularly at the metaor para-positions. Furthermore, the potent activity of analogs 26 (3-phenyl) and 27 (4-phenyl) implies a larger capacity or perhaps flexibility of the putative molecule binding site to accommodate flat hydrophobic substitutions. While the increased lipophilicity is generally helpful to improve BBB permeability it is typically is detrimental to solubility. 55,70 A few analogs that incorporated polar functional groups such as methoxy, hydroxy, and dimethyl amino all resulted in reduced potency against AC1 activity compared to the top leads.
In terms of potency against Ca 2+ /CaM-stimulated AC8 activity, the halogenated series was generally least selective at around 50% AC8 inhibition at the AC1 IC90 with the exception of the 3-Br derivative at 15% AC8 inhibition. When the substitution was switched to lipophilic alkyl functionalities there was an observed drop in AC8 inhibition at the AC1 IC90 for the analogs tested. Interestingly, the branched derivatives we slightly less selective than the linear alkyl counterparts.
Given that the scaffold has three aromatic rings and contains several heteroatoms that could partake in both intramolecular and intermolecular hydrogen-bonds we posited that perhaps the low solubility of the scaffold may be a function of planarity and high-crystal packing energy. 71,72 It has previously been observed that introducing contiguous rotatable bonds may reduce crystal packing and improve solubility of molecules in aqueous media. 71 Thus, analogs 35 and 36 were designed to insert a methylene between the amide carbonyl and the pendent phenyl to explore both AC1 potency and solubility. Unfortunately, the modification was detrimental to AC potency as each analog was at least 10-fold less potent compared to their nearest neighbor analogs without the extra methylene. However, qualitatively these analogs were more soluble in aqueous media compared to the matched molecular pairs, although exact measurements were not obtained because the molecules were inferior in AC1 activity. Nonetheless, for future analog design the strategy of reducing planarity of the scaffold may prove useful. Finally, when combinations of substitutions were made there was not an improvement in activity versus AC1 that was desired. Regardless, the combined modifications in 38 did project to yield a CNS active analog according to QikProp projections. Thus, this molecule, along with three analogs in 16, 20, and 39 that exhibited better potency with inferior projected BBB permeability were prioritized for further evaluation in downstream assays.
In addition to AC1 and AC8, the original hit and four prioritized analogs were also tested against two other AC sub-families represented by AC2 and AC5. AC2 is a member of the group II subfamily and is conditionally activated by G subunits. AC5 is a group III sub-family enzyme and is inhibited by Ca 2+ . It was found that when AC2 is activated by PMA that the scaffold generally potentiates AC2 activity by up to 50%, although analog 38 displayed the relatively little potentiation. This phenomenon has been previously observed for other AC1 inhibitors reported by our group 34,35 and at this point it is difficult to posit what this may mean physiologically, if anything, as there is very little understood about AC2 activity in vivo. AC2 is expressed in the brain, lung, skeletal muscle and heart yet there is very little understood about its physiological function and a lack of animal knockout models and selective inhibitors has kept the function elusive. 3,34 As for AC5 activity the scaffold appeared to have no modulatory effects against this enzyme that differed from the DMSO-treated controls. Therefore, the pyrimidinone scaffold appears to be selective for targeting AC1 over other isoforms.
Little is known about the scaffold at this point from an off-target perspective. The pyrazolepyrimidine substructure has been recently reported in only one other manuscript to possess antileishmania properties via promoting microtubule polymerization. 73 Moreover, three analogs tested for activity against four human kinases relevant to off-target toxicity displayed moderate-to-no inhibition at a single dose of 10 µM. This would suggest that kinase inhibition may not preclude this scaffold from advancement, however, the kinases tested are not exhaustive and additional experiments are necessary to understand the potential impact the molecules may have on the human kinome. Additionally, even though AC8 maintains significant residual activity at higher concentrations of inhibitor we still do not know if this will be sufficient to alleviate potential learning deficits similar to those observed in the AC1 and AC8 double knockout mice. Further analysis in learning and behavior assays are required to fully investigate any potential impact, or lack thereof, on learning and memory. Exploration of SAR at other parts of the scaffold may provide openings for improving BBB permeability. The investigation of this scaffold is only at the beginning stages with much yet to learn about the SAR and potentially how the molecules interact with the target intracellularly that could provide further ideas for analog design. Regardless, the pyrazolo-pyrimidine scaffold represents a previously unreported class of AC1 inhibitors that may have potential for treatment or aid in validation of AC1 inhibition as a viable therapeutic strategy for inflammatory pain and to provide alternatives to opioid analgesics.

CONCLUSION
A cell-based high-throughput screen identified the pyrazolo-pyrimidine scaffold with inhibitory activity against Ca 2+ /CaM-stimulated AC1 cAMP production. A ligand-based SAR optimization campaign, with 36 novel analogs, explored modifications to the benzamide portion of the scaffold to improve AC1 inhibitory activity and improve selectivity over the closely related AC8. It was observed that hydrophobic substitutions on the phenyl ring at both the metaand para-positions provided IC50 values as low as 0.25 µM with reduced activity against AC8.
Prioritized analogs based on potency, selectivity, solubility and predicted BBB permeability were further evaluated at additional AC isoforms. The molecules were observed to potentiate AC2 activity but did not display any modulatory effect against AC5. Additionally, the prioritized analogs were largely inactive at all opioid receptors and against selected kinases. Finally, when evaluated in a murine CFA model for inflammatory pain analog 38 displayed modest antiallodynic properties with a significant effect at the 1-hour post-treatment time point, demonstrating the potential of the scaffold for relieving pain. Further studies will focus on improvement of physicochemical and pharmacokinetic properties while maintaining AC1 potency and selectivity to enhance in vivo efficacy. Compounds were prepared according to scheme 1 and protocol is detailed below for compound 1.
The vial was then heated to 100°C in a Biotage Initiator microwave for 12 hours. After cooling, the mixture was diluted with deionized water and agitated vigorously. The aqueous layer was then     (17). Prepared using general procedure A with 4-methylbenzoic acid (0.040 g, 0.

High-throughput screen for AC1 inhibitors
The inhibitor ST034307 (positive control), or the Life Chemicals library compounds were added at a screening concentration of 10 μM using a pin tool liquid handling system. Following a 30-minute incubation at room temperature with DMSO or the compounds, the calcium ionophore, A23187, was added to all the wells to a final concentration of 3 μM in the presence of the phosphodiesterase inhibitor, 3-isobutyl-1-methylxanthine (IBMX). After 1-hour incubation with A23187 at room temperature, cAMP accumulation was measured using the HTRF cAMP kit from Cisbio (Bedford, MA). The screen was carried out on three separate days and a total of 10,240 compounds were screened. The percentage inhibition (%) was calculated by normalizing the cAMP levels of each of the screen compounds to the mean cAMP levels of the positive control, ST034307, (100% inhibition) and DMSO (0% inhibition) in each plate.

Adenylyl cyclase cAMP accumulation assay
Cryopreserved HEK-ACΔ3/6 cells overexpressing human AC1, AC2, AC5 or AC8 were transferred to a 15-mL Falcon tube and gently resuspended in prewarmed Opti-MEM. 48 The cells were centrifuged for 5 minutes at 150 x g, supernatant was discarded, and the cell pellet was resuspended in 10 mL for a second centrifugation step. Then, the cells were counted, plated in a white opaque 384-well plate and incubated for 1-hour in a 37°C incubator supplemented with 5% CO2. The hit compounds' working solutions were prepared in prewarmed Opti-MEM and successive serial dilutions were prepared using a Precision 2000 automated pipetting system. Dose response curves were generated for each hit compound using a three-fold serial dilution starting at 30 μM concentration (at least 6-points per curve). Compounds were added to the assay plates and incubated for 30 minutes at room temperature. AC activity was selectively stimulated on each AC isoform with 3 μM A23187 (AC1 and AC8), 100 nM PMA (AC2), or a low concentration (1 μM) of the AC activator, forskolin (AC5). All stimulants of AC activity were added in combination with IBMX at a final concentration of 0.5 mM. After 1-hour incubation with the stimulant at room temperature, cAMP accumulation was detected using the fluorescence-based Cisbio HTRF cAMP detection kit.

Thermodynamic solubility assay
The thermodynamic solubility of analogs was determined via an adapted protocol from literature. 53,54 Compounds were dissolved at a concentration of 0.5 mg/mL in DCM (minimum total volume 1.5 mL). 250 µL of this stock were transferred in triplicate to three separate HPLC

MDCK permeability assay
Assay and data analysis were performed by Eurofins Panlabs (MO, USA) according to the following protocol from Hidalgo et al. 77 using MDCKII cells and analysis was carried out as previously described. 78

CYP Inhibition assay
Assay and data analysis were performed by Eurofins Panlabs (MO, USA) according the following protocol adapted from Stresser et al. 79

Human Liver Microsome Stability
Assay and data analysis were performed by Eurofins Panlabs (MO, USA) according to the protocol from Obach et al. 80

Evaluation of compound efficacy in CFA inflammatory pain model
Male and female C57BL/6N mice were obtained from Envigo (Indianapolis, IN). Mice