Discovery of the First Non-cGMP Mimetic Small Molecule Activators of cGMP-Dependent Protein Kinase 1 α (PKG1α)

Abstract

PKG1α is a central node in cGMP signaling. Current therapeutics that look to activate this pathway rely on elevation of cGMP levels and subsequent activation of PKG1α. Direct activation of PKG1α could potentially drive additional efficacy without associated side effects of blanket cGMP elevation. We undertook a high-throughput screen to identify novel activators. After triaging through numerous false positive hits, attributed to compound mediated oxidation and activation of PKG1α, a piperidine series of compounds was validated. The hit 1 was a weak activator with EC₅₀ = 47 μM. The activity could be improved to single digit micromolar, as seen in compounds 21 and 25 (7.0 and 3.7 μM, respectively). Several compounds were tested in a pVASP cell-based assay, and for compounds with moderate permeability, good agreement was observed between the biochemical and functional assays. These compounds will function as efficient tools to further interrogate PKG1α biology.

Cardiovascular diseases and related comorbidities are globally one of the leading causes of human mortality. In the US alone, it is estimated that over the next several years the number of heart failure cases alone will climb to over 8 million patients.¹ While there have been numerous therapeutics developed to treat heart failure, there remains a significant unmet medical need.

Targeting the cyclic guanosine monophosphate (cGMP) signaling pathway has been explored as an option to slow or halt the progression of pathological cardiovascular stress, particularly pulmonary hypertension and heart failure. cGMP is a central second messenger that amplifies (transduces) both the nitric oxide (NO) driven soluble guanylate cyclase (sGC) signaling cascade and the naturetic peptide activated particulate guanylate cyclase (pGC) pathway.² Activation of these pathways leads to elevation of cellular levels of cGMP. cGMP in turn mediates multiple downstream effects on ion channels, phosphodiesterases, and protein kinases.^3,4 One of the protein kinases activated by cGMP in the cardiovascular system is PKG1α (cGMP-dependent protein kinase 1 α).⁵ This serine/threonine kinase consists of a regulatory domain and a catalytic domain that exist as a homodimer.² Within the regulatory domain is an autoinhibitory (AI) subdomain that contains an inhibitory motif called the “pseudo-substrate” motif. The AI, in the absence of cGMP, interacts with the catalytic domain blocking substrate access and thus locking PKG1α in the basal, inactive state. Binding of cGMP to the regulatory domain triggers a significant conformational change that releases the catalytic domain from autoinhibition.⁶

It is also possible to stimulate PKG1α through oxidation of specific cysteine residues and under conditions with high oxidative potential (i.e., H₂O₂).⁷ The activation mechanism of PKG1α has also been studied using X-ray crystallography.⁸⁻¹⁰ Direct activation of PKG1α (vs pharmacological elevation of cGMP levels) could drive additional therapeutic efficacy of the cGMP signaling pathway. Indeed, within the heart, emerging data implicates PKG1α as a critical node and possible therapeutic target to modulate cardiac stress^11,12 and heart failure with preserved ejection fraction.^13,14

Historically, small molecule stimulation of PKG1α was only possible with cGMP mimetics such as 8-bromo-cGMP.^15,16 However, these nucleotide analogues are not selective for PKG1α and also bind many or most of the downstream targets of cGMP with similar or even more potent affinities.¹⁵ Their physicochemical properties also make them challenging to develop as drug candidates. More recently, a set of synthetic peptides have been identified that activate PKG1α in the absence of cGMP.¹⁷ These peptides were derived from the helical segment that bridges the catalytic and regulatory domains of PKG1α. The 20–30 amino acid peptides induce vascular relaxation in vitro using reversible permeabilization to increase the cell permeability of the peptides.¹⁷ In addition to these peptide activators, a set of compounds was identified that induces oxidation of cysteine 42 to activate PKG1α.¹⁸ The hit compound (C1) lowered systemic blood pressure in hypertensive wild-type mice.

In our efforts to discover direct small molecule activators of PKG1α, we undertook a high-throughput screening campaign. As the only known direct activators of PKG1α were either cGMP mimetics or peptides, we anticipated it would be challenging to identify novel chemical matter, as the binding pocket for cGMP is highly constrained. To facilitate the identification of novel activators (i.e., direct or allosteric), the initial screen was conducted with both a partially active PKG1α (EC₂₀ concentration of cGMP was added to the assay) and a “basal” level of activation (no cGMP). However, no hits were identified using the basal state and all subsequent work described herein was conducted using PKG1α in the partially activated state. The objective behind screening in the partially activated state was twofold. The first was to attempt to bias the potential binders away from the cGMP binding pocket. It was hypothesized that compounds that bound in this conserved pocket would have poor selectivity versus other targets of cGMP (as is the case for 8-bromo-cGMP) or would be weak binders. Second, due to the high affinity for cGMP and cAMP for PKG1α,¹⁹ it was believed that the basal concentrations of these PKG1α activators in cells/tissues would lead to most PKG1α being partially activated under physiologic conditions.²⁰

Our strategy was to screen the entire internal compound collection (approximately 2.9 M compounds) to identify activators using the hit identification paradigm summarized in Figure 1. The primary uHTS screen was performed at a single concentration (33 μM, N = 1) using a luminescent-based assay to quantify ADP (ADP-Glo, Promega).²¹ To identify unique hits, statistical analysis metrics, Z* and B-score, were applied. The Z* benchmarks a compound against the median of the test well region on the assay plate. The Z* value for each compound determines how unique a compound is from the rest of the test region. The B-score, on the other hand, accounts for positional biases on the assay plate (that has nothing to do with enzyme activity) which can increase the probability of false positives and false negatives. A preliminary analysis was done using assay plates with DMSO in the test well region to determine the Z* and B-score cutoff criteria. To make sure all possible activators were not missed, any compound that was identified as a hit in either Z* or B-score criteria was included in the primary screen hit list for further validation. To give a sense of the % PKG1α activation of compound hits that made it to the cutoff criteria, hits that made it to Z* cutoff (>4.86) had equivalent % PKG1α activation of >15.5% and hits that made it to the B-score cutoff (>9.74) had equivalent % PKG1α activation of >14.2%. Hits from the primary screen were tested again in triplicate in the Confirmation Screen (single concentration) from stock solutions.

Figure 1.

Screening paradigm to discover PKG1α activators.

The second-tier confirmation screen (ADP-Glo, N = 3) prioritized 4773 primary screen actives to progress to the third-tier confirmation assays for on-target activity. To eliminate false positives due to detection-method-specific compound interference, two orthogonal assays were run in parallel with the primary ADP-Glo assay. A label-free, high-throughput mass spectrometry (HTMS) assay using the RapidFire platform (Agilent) directly tracked the PKG1α-catalyzed phosphorylation of the peptide substrate, Glasstide, while a homogeneous time-resolved fluorescence (HTRF) assay (KinEASE, Cisbio) tracked phosphorylation of a proprietary peptide substrate. The third-tier concentration response assays were performed using an 8-point, 3-fold dilution concentration series starting at 160 μM to ensure coverage of weak activators. Because the HTMS assay is not susceptible to compound-based signal interference, hit prioritization criteria required observation of concentration-dependent activity in the HTMS assay and in at least one orthogonal assay (ADP-Glo or HTRF). In addition, the compounds had to pass a structure alert filter to remove pan assay interference compounds (PAINs).²² Using this approach, 1054 hits were identified, translating to a very low final hit rate of ∼0.0004%.

Upon visual inspection of the chemical matter, a concerning series of compounds was identified, exemplified by pyrimido triazine diamines. These compounds were identified several years ago by teams at Roche²³ and GSK²⁴ and recently covered in a review of frequent promiscuous HTS hits.²⁵ These compounds can function as redox cyclers in the presence of a strong reducing agent (dithiothreitol, DTT) and permit the oxidative formation of hydrogen peroxide in the presence of oxygen.²⁶ PKG1α has cysteine residues susceptible to hydrogen peroxide oxidation, resulting in enzyme activation. Recent work by Sheehe et al.⁷ demonstrated that partial activation of PKG1α by peroxides likely is due to oxidation of cys117. The presence of 1 mM DTT in the screening assay buffer thus created the ideal conditions to pick up redox cyclers as nonspecific (false positive) hits in the primary screen which persisted through the subsequent hit confirmation.

In order to eliminate false positives from the 1054 primary hits, the validation assay buffer was optimized to include a weaker reducing agent, β-mercaptoethanol (BME), to prevent activation of redox cycling and addition of detergent (0.01% Triton X-100) to reduce false positives from soluble aggregation of compounds tested at high concentrations.^6,27 Note that caution in using Triton-X-100 was raised in a recent report due to the possibility of introducing false negatives from suppression of specific compound–target interactions.²⁸

We found that 5 mM BME was sufficient to provide the required reducing environment for PKG1α but a low enough concentration not to support production of reactive oxygen species catalyzed by compound redox cyclers (data not shown).

With these modified assay conditions in hand (5 mM BME, 0.01% Triton X-100), the 1054 hits were retested in the ADP-Glo assay. Surprisingly, only 497 compounds showed activity under these optimized conditions. Indeed, when the chemotypes that now appeared to be inactive in the optimized assay were further investigated for their ability to function as redox cyclers using a H₂O₂-dependent horseradish peroxidase assay, they were found to catalyze the production of H₂O₂ in the presence of a strong reducing agent (data not shown).²⁹ This confirmed that our new assay conditions could selectively remove these false positives. Additionally, all hit classes that remained active under the new assay conditions were also tested in the H₂O₂-dependent horseradish peroxidase assay and were confirmed to have no redox-cycling activity.

Upon further investigation of the remainder of the hits, it was found that many of the chemical classes did not reproduce their activity upon resynthesis or repurification (HPLC) of solid material from our chemical inventory. Upon closer inspection, it was found that the routes used to make many of these compounds utilized a transition metal at the last synthetic step. Transition metal impurities have demonstrated the ability to generate false positives in HTS screening campaigns.³⁰ Specifically for this screen, it is reported in the literature that transition metals can potentially oxidatively dimerize PKG1α.³¹ While we were aware that the oxidative dimerization of PKG1α could lead to activation (vide supra), we did not initially anticipate the presence of transition metal impurities in the screening collection. As the assay was performed at a high top concentration (160 μM), there was a concern that there may be quantities of metal impurities in the compound solutions that could oxidize PKG1α. Indeed, spot checking some of the compound samples identified high micromolar concentrations of palladium, zinc, and copper (data not shown). Accordingly, this prompted further reoptimization of our assay with the addition of EDTA to chelate any trace levels of metals that may contribute to a false-positive assay readout. EDTA alone in the assay buffer was shown to have no impact in the assay. Furthermore, the addition of EDTA was validated to eliminate PKG1a activation observed in the presence of divalent metals (data not shown). As we were confident that both our purification procedure for all new compounds (HPLC) and revised assay conditions with EDTA would remove any undesirable metal mediated effects, there were no further efforts to quantify metal impurities.

With these final, optimized assay conditions in hand, the remaining 497 compounds were rescreened again, resulting in only 36 active compounds. Chemical validation of the structures of these hits (via resynthesis) led to a final set of 24 active hits and 7 hit classes. Out of these actives, two chemical classes were of interest: piperidine 1 and cyclohexyl 2 (Figure 2). The core structures of both 1 and 2 were originally made as part of the discovery work to identify aprepitant (3), a drug developed to prevent chemotherapy induced vomiting that works by blocking NK₁ receptor signaling.^32,33 Aprepitant itself was tested under optimized assay conditions and is not an activator of PKG1α. Cyclohexyl compound 2 demonstrated single digit micromolar potency and was one of the most potent chemical starting points identified. Piperidine 1 was significantly weaker in potency than 2; however, all attempts to identify additional potent analogues of cyclohexyl 2 failed (data not shown). This, combined with overall poor physicochemical properties (solubility, permeability), led to the prioritization of efforts to follow up piperidine 1. However, no structurally similar active hits of 1 were identified from the screen, and a systematic investigation of the minimum pharmacophore was undertaken. Initial work to investigate the stereochemistry of the piperidine ring indicated that the (S,S) configuration was optimal; all other stereochemical permutations were inactive. Additionally, truncation of either the benzyl ether or the pendant phenyl ring led to loss of activity. Surprisingly, when the core piperidine ring was changed to cyclohexyl (as in 2) again the analogues were inactive. Small modifications of the dichlorophenyl ring (removal of Cl, introduction of heteroatoms, etc.) also led to a loss of activity. Lastly, replacement of the ether oxygen with carbon or nitrogen also led to inactive analogues. It seemed apparent that the substituents and configuration of the piperidine ring were critical for activity and deviations were not tolerated.

Figure 2.

Validated PKG1α activator hits.

Due to the challenges in modification of the core, attention turned to modification of the periphery of piperidine 1 to improve the potency of this weak hit. N-Benzyl substituents were explored first (Table 1). Methylation of the imidazole ring appeared to be tolerated (compound 4), but other 5-membered heterocycles such as thiazole 5 and oxazole 6 resulted in loss of activity. N-Methyl pyrazole 7 showed modest activation. Investigation of 6-membered rings showed that neither phenyl (8) nor 2-pyridyl (9) were active. 3-Pyridyl (10) did show activity, but moving the nitrogen to the 4-position (11) again lost activity. Interestingly, when a carboxylic acid was introduced to the 3-pyridyl to generate compound 12, the compound was equipotent to the parent 3-pyridyl (10). Moving the carboxylic acid adjacent to the pyridyl nitrogen resulted in a loss of activity (compound 13). When the pyridyl nitrogen was removed from compound 13 and the carboxylic acid retained to generate compound 14, the activity returned. This activity was retained when the carboxylic acid was switched to a tetrazole ring (15). Additional modifications to the phenyl by combining a 2-methoxy group with a carboxylic acid retained modest potency (16 and 17). When the carboxylic acid was combined with a 5-membered heterocycle (e.g., thiophene 18), potency could still be retained. At this stage, while we were able to show that several different substituents were tolerated at the N-benzyl position, none were able to improve the potency over the parent hit piperidine 1.

Table 1. Modifications of the N-Benzyl Group.

^aEC₅₀ values are the average of two to five independent trials using the ADP-Glo assay format. Please see the Supporting Information for tabulated values with errors expressed as standard deviations.

To identify a path forward for the piperidine series, a more systematic exploration of the structure–activity relationships (SARs) of the core was conducted. During these studies, it was found that truncation of the N-benzyl group to N-methyl resulted in a loss of activity (piperidine 19, Figure 3). However, this same truncation was tolerated when a hydroxyl methyl group was added to the benzylic ether to give compound 20. With this discovery in hand, the next step taken was to reinvestigate the SAR of the N-benzyl position to further improve the potency.

Figure 3.

A hydroxyl methyl group could improve potency, as revealed by truncation studies.

Exploration of the N-benzyl position on the hydroxyl methyl core is presented in Table 2. An immediate improvement in the potency was evident; compound 21 is the hydroxyl methyl variant of the original piperidine hit 1, and potency was improved almost 7-fold. Likewise, improvements in the potency were observed for the 3-pyridyl carboxylic acid 22 and 4-carboxylic acid 23 (compare to 12 and 14, respectively). However, improvements in potency were not uniform, as the potency of tetrazole 24 was unchanged on this core. A more thorough exploration of structural diversity showed that pendant heterocycles (25) and fused heterocycles (26, 27) were tolerated. These active compounds provide some promising avenues to further expand the SAR and improve on the potency of these activators.

Table 2. Benzyl Modifications on the Hydroxyl Methyl Core.

^aEC₅₀ values are the average of two to five independent trials using the ADP Glo assay format. Please see the Supporting Information for tabulated values with errors expressed as standard deviations.

To further assess the potential developability of this series of PKG1α activators, additional profiling of key compounds was conducted including measurement of the aqueous solubility, permeability, and cell-based activity. The data is presented in Table 3. Activity in a whole cell assay was deemed critical to demonstrate the functional effects of the identified compounds as validated activators of PKG1α. Vasodilator-stimulated phosphoprotein (VASP) is a well characterized substrate of PKG1α and is phosphorylated by PKG1α on serine 239.^34,35 As aqueous solubility and cell permeability would influence the activity, these compound properties were also monitored. In general, all of the compounds tested had excellent solubility with most of the compounds having >100 μM solubility at pH 7. However, the cell permeability of the compounds of interest was more variable. Several had poor permeability (defined as <10 × 10^–6 cm/s) including compounds 1 and 27. This poor solubility trended with a significant cell potency shift or loss of activity. In contrast, compounds that displayed a more moderate permeability ((10–20) × 10^–6 cm/s) such as compounds 21 and 23 showed almost no shift, and cell-based pVASP activity was in good agreement with the ADP Glo derived biochemical assay.

Table 3. Physicochemical Properties and Cell-Based Activity of Key Compounds.

compound	pH 7 solubilitya (μM)	MDCKII Papp (×10–6 cm/s)b	pVASP (ser 239) cell assay (μM)
1	120	2	>164
2	154	15	12
21	60	19	8.8
22	74	14	28
23	85	16	15
26	154	11	141
27	128	4	140

^aSee the Supporting Information for the assay protocol.

^bPermeability measured according to a published procedure.³⁶

The general synthesis of this piperidine class of PKG1α activators is presented in Schemes 1 and 2. The synthesis of the Boc-protected piperidine core 28 has been reported previously.³⁷ Alkylation of 28 with benzyl chloride 29 followed by Boc deprotection yielded piperidine ether 30. Reaction of piperidine ether 30 with protected imidazole mesylate 31(38) followed by deprotection with TFA provided piperidine 1.

Scheme 1. Synthesis of Piperidine 1.

Reagents and conditions: (a) NaH, DMF, rt, 79%; (b) TFA, DCM, rt, 100%; (c) K₂CO₃, 31, DMF; (d) TFA, DCM, 45% over two steps.

Scheme 2. Synthesis of Hydroxyl Methyl Piperidine 21.

Reagents and conditions: (a) Rh₂(OAc)₄, benzene, reflux, 41%; (b) TFA, DCM, rt, 100%; (c) K₂CO₃, 31, DMF; (d) TFA, DCM, 40% over two steps; (e) LiBH₄, MeOH, chiral resolution, 10%; (f) DBU, 89%.

The synthesis of the hydroxyl methyl piperidine core is presented in Scheme 2. Reaction of piperidine 28 with diazo ester 32 generated the hydroxyl methyl piperidine core 33 after Boc deprotection. Alkylation of the piperidine nitrogen with mesylate 31 followed by deprotection gave compound 34. Reduction and chiral resolution generated hydroxyl methyl piperidine 21 in modest yield.

In summary, we discovered the first non-cGMP mimetic small molecule activators of PKG1α. These compounds were discovered using a high-throughput screen with a very low hit rate. During the hit workup, multiple false positive hits were identified and resulted from either redox cycling or transition metal impurities that led to oxidative activation of PKG1α. After adjusting the assay conditions to eliminate these compounds, a weakly active hit was identified, exemplified by compound 1. Preliminary SAR of the N-benzyl position identified several additional active compounds but no real breakthroughs in potency. Truncation studies revealed a hydroxyl methyl group as the potential path forward to improving the PKG1α activation of the series. Indeed, reinvestigation of the N-benzyl position showed that single digit micromolar potency was achievable. Several active compounds were followed up in a pVASP cell-based functional assay which showed that compounds that had moderate solubility and permeability had comparable potencies between the two assays, lending additional confidence to the hit. This series of compounds shows promise as tools to further interrogate PKG1α biology and as a potential avenue to drive additional efficacy from this signaling pathway. Future studies will focus on further optimization of this class of activators to test in models of cardiovascular disease as well as to understand the nature of the binding to PKG1α and mechanism of activation.

Glossary

Abbreviations

cGMP

cyclic guanosine monophosphate

NO

nitric oxide

sGC

soluble guanylate cyclase

pGC

particulate guanylate cyclase

PKG1α

cGMP-dependent protein kinase 1 α

uHTS

ultrahigh-throughput screen

ADP

adenosine diphosphate

HTMS

high-throughput mass spectrometry

HTRF

homogeneous time-resolved fluorescence

PAINs

pan assay interference compounds

DTT

dithiothreitol

BME

β-mercaptoethanol

SAR

structure–activity relationships

VASP

vasodilator-stimulated phosphoprotein

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsmedchemlett.1c00264.

• Representative synthetic procedures, compound characterization, and assay protocols (PDF)

Author Present Address

^∇ Janssen R&D, Spring House, Pennsylvania 19477, USA

Author Present Address

^○ Schrodinger, New York, New York 10036, USA.

Author Contributions

^$ J.H. and A.S. contributed equally to the work described in this manuscript. All authors have given approval to the final version of the manuscript.

The authors declare the following competing financial interest(s): At the time the work described herein was completed, all authors were employees of Merck & Co., Inc. Kenilworth NJ.