Discovery of PLD4 modulators by high-throughput screening and kinetic analysis

Abstract

Phospholipase D3 (PLD3) and D4 (PLD4) are endolysosomal exonucleases of ssDNA and ssRNA that regulate innate immunity. Polymorphisms of these enzymes are correlated with numerous human diseases, including Alzheimer’s, rheumatoid arthritis, and systemic sclerosis. Pharmacological modulation of these immunoregulatory proteins may yield novel immunotherapies and adjuvants. A previous study reported a high-throughput screen (N = 17,952) that discovered a PLD3-selective activator and inhibitor, as well as a nonselective inhibitor, but failed to identify selective modulators of PLD4. However, modulators selective for PLD4 are therapeutically pertinent, since recent reports have shown that regulating this protein has direct implications in cancer and autoimmune diseases. Furthermore, the high expression of PLD4 in dendritic and myeloid cells, in comparison to the broader expression of PLD3, presents the opportunity for a cell-targeted immunotherapy. Here, we describe screening of an expended diversity library (N = 45,760) with an improved platform and report the discovery of one inhibitor and three activators selective for PLD4. Furthermore, kinetic modeling and structural analysis suggest mechanistic differences in the modulation of these hits. These findings further establish the utility of this screening platform and provide a set of chemical scaffolds to guide future small-molecule development for this novel immunoregulator target.

Graphical Abstract

1. Introduction

The inflammation produced by pro-inflammatory cytokines and type I interferons are triggered by nucleic acid sensors.^1,2 These sensors, such as toll-like receptors (TLR), reside in endosomes and recognize double-stranded RNA (TLR3), single-stranded RNA (TLR7 and TLR8), single-stranded DNA containing unmethylated CpG motifs (TLR9), and bacterial 23S RNA (TLR13).^3–8 Dysregulation of these TLR can initiate or aggravate pathogenesis.^8,9 Thus, TLRs and their signaling pathways have been targeted as potential therapeutics against numerous ailments, such as cancer, autoimmunity, and inflammation.^1,10–19 Such studies have emphasized the importance of modulating innate nucleic acid sensors as novel immunotherapies.

Phospholipase D3 (PLD3) and D4 (PLD4) are endolysosomal exonucleases that regulate innate immunity by digesting the single-stranded ligands of nucleic acid sensors.^20,21 PLD isoforms exhibit structural homology; however, PLD3 and PLD4 do not share the same catalytic activity of PLD1 and PLD2.^22–25 Thus, PLD family members demonstrate distinct functions and intracellular localizations.^20,26–28 One such distinction is the immunoregulator function of PLD3 and PLD4, which was established through the correlation of their polymorphisms with autoimmune and inflammatory diseases.^22,29–39 Modulating PLD3 and PLD4 may provide therapeutics to their associated diseases, serve as adjuvants, and act as immune suppressants/stimulants.

A previous study from our laboratory established a platform for a high-throughput enzymatic screen to discover modulators of PLD3 and PLD4 using a 3’ fluorogenic mononucleotide substrate, 3’-dT-MUP.⁴⁰ Through this endeavor, a PLD3 activator, PLD3 inhibitor, and nonselective inhibitor were discovered—validating our approach. However, none of the hits were selective for PLD4, which is implicated in distinctly different disease states from PLD3. Thus, in this study, we conducted a screening campaign of a diversity library (N = 45,760), where we identified PLD4-selective inhibitors and activators. These hits were implemented in potency and kinetic modulation assays to characterize their activity. With the elucidation of a small-molecule specific for PLD4 modulation, this protein’s biological role can be further probed in associated disease pathways.

2. Materials and Methods

2.1. Reagents and materials

Synthesis reagents were purchased from Millipore Sigma. Black half-area 96-well plate (Corning^® 3694). 384-well black opaque plate (Corning^® 3573). The diversity library was provided by Luke Lairson and is comprised of compounds from Life Chemicals and ChemDiv. Hits were purchased from either Life Chemicals or ChemDiv. PLD3 and PLD4 were provided by David Nemazee, Ph.D. and Linghang Peng following previously published protocols.²⁰

2.2. Equipment

EnVision Plate Reader (PerkinElmer). General bs50/bs50 Dual Mirror (PerkinElmer, #651). Optical Filter Umbelliferone 355 (PerkinElmer, #103). Optical Filter FITC 535 (PerkinElmer, #206). Bravo Automated Liquid Handling Platform (Agilent).

2.3. Chemistry - Synthesis

The substrate (3’-dT-MUP) was synthesized according to previously published procedures.⁴⁰ The synthetic scheme, experimental methods, and characterization for the substrate can be found in the supporting information of prior publications.⁴⁰

2.4. High-throughput screen

Assay buffer (50 mM NaOAc, 0.03% Tween-80, 0.67 mM EDTA, 0.002% BSA, pH 5.0) was used in all screening and hit validation studies. A positive control plate was made for PLD3 or PLD4 screening by transferring positive control compounds (A1 for PLD3 and I2 for PLD4) detailed in a previous publication to column 23, row A-H of a 384-well black plate (Corning^® 3573).⁴⁰ The exploratory screen (1% DMSO in assay buffer) and kinetic analyses (2% DMSO in assay buffer) were conducted in singlicate, while hit validation (2% DMSO in assay buffer) and potency (2% DMSO in assay buffer) were performed in triplicate.

2.4.1. Screen

The screening libraries and positive control plates were dispensed (Bravo Automated Liquid Handling Platform, 100 nL, conc_final = 10 μM) to 384-well black plates (Corning^® 3573) with either PLD3 (10 μL, conc_final = 0.5 nM) or PLD4 (10 μL, conc_final = 1 nM). After the compounds and enzymes were incubated (1 h, 21 °C), 3’-dT-MUP substrate (10 μL, conc_final = 20 μM) was added to plates. The fluorescence intensity was read as described vide infra.

2.4.2. Counterscreen

The screening libraries and positive control plates were dispensed (Bravo Automated Liquid Handling Platform, 100 nL, conc_final = 10 μM) to 384-well black plates (Corning^® 3573) with interference compound (4-methylumbelliferone, 20 μL, conc_final = 5 μM). After the compounds and 4-methylumbelliferone were incubated (1 h, 21 °C), fluorescence intensity was read as described vide infra.

2.4.3. Fluorescence intensity measurement

All fluorescence intensity (FI) measurements for 384-well plates (Corning^® 3573) were performed on an EnVision plate reader (PerkinElmer) with 20 μL assay solutions per well. All FI measurements for 96-well half-area plates (Corning^® 3694) were performed on an EnVision plate reader (PerkinElmer) with 50 μL assay solutions per well for hit validation and 40 μL assay solutions per well for potency and modulation analyses.

2.4.4. Screen analysis

FI emission values were converted to slope (Δemission/Δtime, RU/s) for plates containing enzyme, while FI emission values were averaged for plates containing interference compound. Percent response was obtained by using the slope or averages in the following equation:

$$\text{Percent}\text{Response}=100\times \left(\frac{\text{Test}\text{Well}-\text{Mean}\text{Low}\text{Control}}{\text{Mean}\text{High}\text{Control}-\text{Mean}\text{Low}\text{Control}}\right)$$

For screening plates, the mean high control (enzyme and substrate only) was determined from 8 replicates and the mean low control (buffer and substrate only) was determined from 16 replicates. Positive controls were assessed in 8 replicates as internal controls. For counterscreen plates, the mean high control (4-methylumbelliferone only) was determined from 8 replicates and the mean low control (buffer only) was determined from 16 replicates. Screening statistics and hit cutoffs were calculated as previously described.^40,41

2.5. Hit analyses

Structure similarity analysis was performed by DataWarrior 5.5.0 with a 80% similarity limit.⁴² 357 Life Chemicals and 312 ChemDiv hits were selected for filtering by the Free ADME-Tox Filtering Tool (FAF-Drugs4, Table S1).⁴³ Classification of compounds from filtering are detailed in Table 1. 35 compounds were purchased from either Life Chemicals or ChemDiv (Table S2) and purity was confirmed by LC-MS. Compounds were dissolved in DMSO as 15 mM solutions. Hits were validated in triplicate using black half-area 96-well plates (Corning^® 3694) and screening conditions/analysis described vide supra. Activity was confirmed for 14 compounds.

Table 1. Results of FAF-Drugs4 filtering of hits.

Hit compounds (N = 669) were assessed through the Free ADME-Tox Filtering Tool (FAF-Drugs4) with the drug-like filter parameter.

	Accepted			Intermediate			Rejected
Type	Life Chemicals	ChemDiv	TotaM	Life Chemicals	ChemDiv	Total 1	Life Chemicals	ChemDiv	Total
Activator	93	92	185	1	4	5	1	4	5
Inhibitor	246	196	442	6	3	9	10	13	23
Total	339	288	627	7	7	14	11	17	28

2.6. Potency determination

The 14 confirmed hits were evaluated for IC₅₀ or EC₅₀ values using a seven-point dilution series (1:2, 300 – 0.4 μM). Potencies were evaluated in triplicate in black half-area 96-well plates (Corning^® 3694) using a previously published method.⁴⁰ Potency values were determined by fitting the triplicate dose response data in GraphPad Prism 10.0.0 following previously published protocols.⁴⁰ Hits with potencies higher than 50 μM were excluded from further analyses.

2.7. Kinetic analyses

Enzyme activity was measured in black half-area 96-well plates (Corning^® 3694) by matrix of 3’-dT-MUP (eight-point dilution, 1:1.5, 300 μM – 0.49 μM) and compound (nine-point dilution, 1:1, 125 μM – 0.49 μM). Velocity (RU/s) was determined by converting the FI emission values to slope (Δemission/Δtime). The data points were fit to kinetic models of inhibition (competitive, uncompetitive, noncompetitive, or mixed) or activation (uncompetitive, noncompetitive, or mixed)⁴⁴ that are shown in Scheme S1 using GraphPad Prism 10.0.0 with a least squares regression fitting method. The best model was selected following previously published methods.⁴⁰

3. Results and Discussion

3.1. Exploratory high-throughput screen.

In order to identify activators and inhibitors of either PLD3 or PLD4, we implemented an enzymatic assay using a 384-well format.⁴⁰ The enzymatic screen used 10 μM of compound, 20 μM of fluorescent substrate (3’-dT-MUP), and either 0.5 nM of PLD3 or 1 nM of PLD4. For internal positive controls, we incorporated previously reported A1 into PLD3 assays and I2 into PLD4 assays (Figure 1).⁴⁰ These controls were chosen based on their potent nature. To reduce the number of false positives due to fluorophore interactions with screened compounds, we included a counterscreen consisting of 10 μM of compound and 5 μM of 4-methylumbelliferone.

Figure 1.

Substrate and positive controls used in the screen. The substrate is a 3’ isomer of thymidine 4-(4-methylumbelliferone) phosphate (3’-dT-MUP). The PLD3 positive control is a previously reported activator (A1). The PLD4 positive control is a previously reported mixed inhibitor (I2).

Both enzymatic screens and counterscreens were read for fluorescence intensity in kinetic modes. For screens containing enzyme, the emissions were converted to slope measurements and inserted into a mathematical algorithm with the low control (substrate) set at 0% and high control (enzyme and substrate) set at 100%, to obtain the relative percent response of wells containing compound (see Materials and Methods). Counterscreen emissions were averaged and inserted into the same mathematical algorithm with the low control (buffer) set at 0% and high control (4-methylumbelliferone) set at 100%. The lower and upper limit percent response for determining hits from the enzymatic screen was 50% and 150%, respectively. Counterscreen cut-offs were calculated by calculating ± 3 standard deviations from the high control (Figure 2). The exploratory screen of a diversity library (N = 45,760) exhibited hits rates for PLD3 (1.27%), PLD4 (0.52%), and counterscreen (14.45%) that were consistent with previous exploratory screens.⁴⁰ A high average Z’ score for PLD3 (0.83 ± 0.05), PLD4 (0.83 ± 0.03), and counterscreen (0.88 ± 0.02) were observed (Figure 2D).

Figure 2.

Exploratory screen for PLD3 and PLD4 modulators. Scatter plots of screening results for PLD3 (A), PLD4 (B), and counterscreen (C) against a diversity library (N = 45,760). Each data point represents the activity (percent response) of compounds (black), positive controls (blue), high controls (green), or low controls (red). Screening statistics are shown (D). The Z’ values are represented as means ± standard deviation. The upper and lower limits show the percent response cut-off values. The hit rates are written in parentheses next to the number of hits.

Abbreviation: PLD, phospholipase D; ctrl, control.

3.2. Structural analyses of hits.

Of the 820 total hits for both enzymes, 669 were available for purchase. These compounds were compiled and assessed through the Free ADME-Tox Filtering Tool (FAF-Drugs4) using a drug-like filter (Table S1).^43,45 The filter classified compounds as accepted, intermediate, or rejected. Hits were considered intermediate due to their status as pan-assay interference compounds or were rejected because of rigid bonds. Of the 195 activators, the filter accepted 185, considered 5 as intermediate, and rejected 5. Of the 474 inhibitors, 442 were accepted, 9 were intermediate, and 23 were rejected (Table 1).

The 190 activators and 451 inhibitors not rejected by filtering, were compiled into DataWarrior 5.5.0 for similarity analysis to identify common chemotypes. Clustering of combined hits resulted in 8 chemotype structures, which were surprisingly evenly categorized by 4 inhibitor and 4 activator chemotypes (Figure 3). Similar to previously disclosed chemotypes, C5 and C6 contain pyrimidine core motifs that mimic the enzyme’s natural substrate structure.⁴⁰ The two chemotypes that share the majority of hit clusters, C1 and C2, contain pyridazine scaffolds, which have been reported to possess a broad-spectrum of pharmacological activities related to mechanisms like DNA damage.^46–48 Furthermore, partial pyridazine analogues have displayed potential applications for immune regulation and epigenetics.^47,48 In order to purchase compounds representative of chemotype clusters and singleton hits, we purchased 6 activators and 29 inhibitors from Life Chemicals and ChemDiv, where 7 compounds comprised of chemotype clusters C1, C2, C3, and C8 (Table S2).

Figure 3.

Chemotype clusters for hits against PLD3 and PLD4. Red represents inhibitor chemotypes and green represents activator chemotypes. Structure similarity analysis was performed by DataWarrior 5.5.0 with an 80% similarity limit.

3.3. Potency analyses of hits.

The 35 compounds were dissolved in DMSO at 15 mM solutions and tested in triplicate at final concentrations of 10 μM using the enzymatic and counterscreen assay in 96-well format. There were 13 hits that interfered with the fluorophore in the counterscreen, so they were excluded from further analyses. The remaining 22 compounds were implemented in potency analyses to determine IC₅₀ and EC₅₀ values. While testing the hits in triplicate, only 4 compounds exhibited potency less than 50 μM (Table 2). Moreover, these hits were all selective for PLD4. I3 displayed strong similarity to the C3 chemotype cluster, while A4 exhibited a strong similarity to C8.

Table 2. Structures and potencies of selected hits.

All compounds are numbered and are prefixed with an “I” for inhibitors and “A” for activators. IC₅₀ and EC₅₀ values are the triplicate means ± SEM derived from fitting a four-parameter sigmoidal dose-response curve of a 7-point dose-response titration (1:2, 300 μM - 0.4 μM) against PLD4.

Compound	I3	A2	A3	A4
Potency (μM)	39.6 ±4.6	35.7 ±4.2	20.2 ±3.3	38.6 ± 9.9

3.4. Kinetic analyses of hits.

All four PLD4 hits were analyzed to understand the modulation relationship between substrate and compound. Change in enzyme activity was determined under steady-state conditions by preparing a matrix with an 8-point dilution of 3’-dT-MUP and 9-point dilution of compound. Enzyme inhibition or activation was fit to various models and the extra sum-of-squares F test was used to determine the most representative model, where improvements in the sum of squares were weighed against the loss of degrees of freedom. PLD4 inhibition by I3 was fit to competitive, uncompetitive, noncompetitive, and mixed inhibition models (Scheme S1). Enzyme inhibition by I3 fit a competitive model (α ≫ 1) with positive cooperativity (Hill coefficient = −1.40 ± 0.19, Figure S1). The selectivity and structural definition of PLD4 outlined in previous publications may explain why PLD4 inhibition fits the competitive model.²⁰ This result is consistent with a previously disclosed non-selective inhibitor.⁴⁰

Enzyme activation by A2, A3, and A4 was fit to uncompetitive, noncompetitive, and mixed activation models (Scheme S1).^40,44 Activation of PLD4 by A2, A3, and A4 all fit the more complex mixed model best (P < 0.0001, Figure 4). All three activators preferentially bound the enzyme-substrate complex (α < 1) with positive cooperativity (Hill coefficient > 1, Figure S1). For mixed models of activation, the activator binds the enzyme and enzyme-substrate complex, thus modifying the dissociation constant (K_m) and rate constant (k_cat) by α and β, respectively. Based on these adjustments, A2 reduced the K_m by 1.7-fold and increased k_cat by 2.4-fold, A3 reduced the K_m by 1.7-fold and increased k_cat by 2.3-fold, and A4 reduced K_m by 1.1-fold and increased k_cat by 1.5-fold.

Figure 4.

Kinetic models for modulation of hits. Substrate digestion was measured in singlicate by matrix of 3’-dT-MUP (eight-point dilution series, 1:1.5, 300 μM – 0.49 μM) and compound (nine-point dilution series, 1:1, 125 μM – 0.49 μM). Data was fit to kinetic models of inhibition or activation (Scheme S1) and the extra-sum-of-squares F test was applied to determine the most representative model. The table contains best-fit values ± standard error.

Three of the selected hits (I3, A2, and A3) all possess a thiophene that is also present in the C3 chemotype. Thiophene-containing compounds possess antimicrobial activity and prior studies have revealed the mechanism of action for such molecules relied on the stabilization of DNA-cleavage complexes for either single or double stranded DNA by selectively binding to the protein.^49,50 Based on these findings, it is possible to speculate on the mechanism of modulation for the potent hits, where I3, A2, and A3 may potentially operate through a similar mechanism of other thiophene-containing moieties by modulating the enzyme-substrate complex of PLD4 and single-stranded nucleic acids, which is supported by our kinetic analyses.

4. Conclusions

PLD3 and PLD4 have been associated with various human ailments, such as Alzheimer’s and autoimmune diseases.^{29,31,32,51–53} Recent studies have demonstrated that these phospholipases are 5’ single-stranded DNA and single-stranded RNA exonucleases that are critical for regulating TLR and cytoplasmic nucleic acid sensing pathways.^20,21 While studies are ongoing to further elucidate their roles in human disease, effective chemical probes are crucial to assist in advancing the comprehension of PLD3 and PLD4 dysregulation in various pathways.

In summary, we implemented a high-throughput enzymatic screen against a diversity library (N = 45,760) to discover inhibitors or activators of PLD3 and PLD4. Prior research had discovered PLD3 activators and inhibitors, both selective and nonselective in nature.⁴⁰ From the hits screened in this study, we identified one inhibitor and three activators that were selective for PLD4. The inhibitor may be therapeutically relevant for cancer, since a study found PLD4 inhibition reduced pro-inflammatory cytokines in macrophages connected to colon cancer.⁵⁴ The activators identified in this study could have potential therapeutic value in treating rheumatoid arthritis and systemic sclerosis, since PLD4 has been found to be deficient in humans with these diseases.^51,52,55 Furthermore, murine studies have shown that PLD4 deficiency can cause a diverse range of rheumatological abnormalities.^20,39 Establishing the structure-activity relationship for these hits will optimize their potency/selectivity and may aid in elucidating the mechanism of action. These hits, and the screening platform that discovered them, have the potential to provide novel therapeutics and means to explore immunoregulator pathways.

Abbreviations

PLD3

Phospholipase D3

PLD4

phospholipase D4

3’-dT-MUP

3’−4-(4-methylumbelliferone) phosphate

FAF-Drugs4

Free ADME-Tox Filtering Tool