Discovery and optimization of piperazine-1-thiourea-based human phosphoglycerate dehydrogenase inhibitors

Proliferating cells, including cancer cells, obtain serine both exogenously and via the metabolism of glucose. By catalyzing the first, rate-limiting step in the synthesis of serine from glucose, phosphoglycerate dehydrogenase (PHGDH) controls flux through the biosynthetic pathway for this important amino acid and represents a putative target in oncology. To discover inhibitors of PHGDH, a coupled biochemical assay was developed and optimized to enable high-throughput screening for inhibitors of human PHGDH. Feedback inhibition was minimized by coupling PHGDH activity to two downstream enzymes (PSAT1 and PSPH), providing a significant improvement in enzymatic turnover. Further coupling of NADH to a diaphorase/resazurin system enabled a red-shifted detection readout, minimizing interference due to compound autofluorescence. With this protocol, over 400,000 small molecules were screened for PHGDH inhibition, and following hit validation and triage work, a piperazine-1-thiourea was identified. Following rounds of medicinal chemistry and SAR exploration, two probes (NCT-502 and NCT-503) were identified. These molecules demonstrated improved target activity and encouraging ADME properties, enabling both in vitro and in vivo assessment of the biological importance of PHGDH, and its role in the fate of serine in PHGDH-dependent cancer cells.


Introduction
Serine is an important biochemical building block, being incorporated wholly into proteins as well as into DNA and RNA bases as one-carbon units or as serine-derived glycine. Given this crucial role, proliferating cells obtain serine exogenously or derive it from glucose. By catalyzing the first, rate-limiting step in the synthesis of serine from glucose, phosphoglycerate dehydrogenase (PHGDH) controls flux through this biosynthetic pathway. A series of PHGDHoverexpressing breast cancer cell lines display a dependence on this pathway as genetic suppression of PHGDH was shown to be toxic even in the presence of exogenous serine. [1][2] Given PHGDH's potential as a drug target in certain cancers, a number of PHGDH inhibitors have been reported. [3][4][5][6][7][8][9][10][11] AstraZeneca used a fragment-based lead discovery approach followed by medicinal chemistry optimization to develop an indole derivative ( Figure 1) with submicromolar K D . [4][5] Cantley and colleagues used a biochemical high-throughput screen to identify a PHGDH inhibitor (CBR-5884, Figure 1) with an IC 50 of ~33 μM. 6 A patent filed by Raze Therapeutics disclosed a chemotype (representative example shown in Fig 1), but no detailed information on inhibitory activity was shown. 9 We have previously reported inhibitors of PHGDH (NCT-502, NCT 503, Figure 1) with lowmicromolar biochemical inhibition that demonstrated inhibition of PHGDH-dependent cancer cell growth, and reduced glucose-derived serine production in cells. 8 Activity was also demonstrated in orthotopic xenograft tumors.
Here, we describe the development and optimization of a high-throughput amenable PHGDH biochemical activity assay to enable screening in 1536-well format. This assay was utilized to conduct a quantitative high-throughput screen (qHTS) to uncover small molecule inhibitors of PHGDH for further development. In addition, we describe the optimization of a thiourea chemotype into a set of in vitro and in vivo probes to study PHGDH biology.

Results
The optimized PHGDH assay used for the HTS utilized a coupled enzyme system to provide robust recombinant PHGDH activity and to shift detection of dehydrogenase activity from a classic NADH-based blue fluorescent readout to a red-shifted resorufin-based readout. This coupled biochemical assay system includes two human enzymes -phosphoserine transaminase (PSAT1) and phosphoserine phosphatase (PSPH) -which naturally participate in the serine biosynthetic pathway immediately downstream of PHGDH, along with a bacterial enzyme, diaphorase from Clostridium kluyveri, to couple NADH levels to the resazurin/resorufin fluorescent dye system (Figure 2a). 12

Assay Development & Optimization
The coupling efforts detailed here were undertaken after initial adaptation of a PHGDH assay containing solely the dehydrogenase and its substrates, which quantified NADH production by directly measuring NADH fluorescence, yielded low enzymatic turnover and required a lengthy reaction period to achieve significant signal (data not shown). Furthermore, given the prevalence of blue autofluorescence among small molecules in screening collections, we anticipated a need to shift the primary assay away from the blue NADH-based fluorescent readout. 13 To address these issues, we coupled NADH production to diaphorase, which has been utilized in other dehydrogenase and NAD + /NADH systems to introduce a separate, red-shifted detection dye, resazurin. 14 Diaphorase is an oxidoreductase enzyme that can utilize NADH or NADPH to catalyze the reduction of resazurin to fluorescent resorufin. Diaphorase-coupling provides a twofold benefit in the case of PHGDH by 1) generating resorufin, a robust detection dye with HTSamenable red-shifted fluorescence, with 1:1 stoichiometry to NADH, and by 2) regenerating NAD + and preventing accumulation of NADH by the primary PHGDH reaction (which can slow enzyme kinetics). We tested Clostridium kluyveri diaphorase for this purpose by determining optimal resazurin and diaphorase concentrations using PHGDH assay buffer and substrate conditions. Titration of resazurin using 0.5 mM NADH and 0.015 mg/mL diaphorase yielded a K m of 50 μM for resazurin, and determined that maximal production of resorufin signal occurred at 250 μM substrate; concentrations above this threshold were found to demonstrate feedback inhibition with reduced signal (Figure 2b). For coupling purposes, resazurin was included at 0.1 mM, which was high enough to achieve a non-limiting concentration relative to NAD + /NADH in the enzyme system, yet low enough to avoid any potential feedback inhibition.
PHGDH has been shown to be susceptible to feedback inhibition 15 by its product phosphohydroxypyruvate (p-Pyr), so the assay system was further coupled to two downstream enzymes, PSAT1 and PSPH, to minimize p-Pyr accumulation and help 'drive' the reaction forward. Optimal stoichiometry of PHGDH, PSAT1 and PSPH were determined by titrating the downstream enzymes PSAT1 and PSPH in the presence of 10 nM PHGDH and non-limiting substrate concentrations (1 mM NAD + , 2 mM 3-phosphoglycerate [3PG], 0.625 mM glutamate), using NADH fluorescence as a direct readout of biochemical reaction progression. Inclusion of PSAT1 and PSPH led to increased PHGDH turnover and greater assay signal at all tested concentrations and stoichiometries, though PSAT1 concentrations were found to most strongly influence overall assay signal ( Figure 2c). Inclusion of 500 nM PSAT1 and 500 nM PSPH yielded a 36% increase in signal over the lowest concentrations tested (100 nM each); though higher PSAT1 and PSPH concentrations did provide further increases in signal. Therefore, 500 nM concentrations of both enzymes were chosen to balance assay signal with protein requirements.
Next, we performed a diaphorase titration in the presence of either PHGDH alone, PHGDH with PSAT1 and PSPH (positive control), or PSAT1 and PSPH alone (no PHGDH, negative control).
The fully-coupled PHGDH/PSAT1/PSPH/diaphorase assay conditions yielded significantly greater assay signal and turnover than other conditions, with 2-7-fold increased turnover compared to PHGDH/diaphorase in the absence of PSAT1 and PSPH (Figure 2d), confirming that the coupled assay increases PHGDH enzymatic activity. Though maximal assay signal was seen at 0.025 mg/mL diaphorase in the fully-coupled conditions, diaphorase was included in subsequent screening at a higher concentration (0.1 mg/mL) to buffer against any significant false positive readings in the event of direct inhibition of diaphorase itself.
With coupling conditions determined, the assay system substrates 3PG, NAD + (PHGDH) and glutamate (PSAT1) were individually titrated to determine K m values and saturating concentrations. K m values for 3PG and NAD + were determined to be 50 μM and 150 μM, respectively, and these concentrations were adopted into the final assay conditions to facilitate detection of small molecule competitors of either substrate (Figures 2e, f). Glutamate was found to be non-limiting above 310 μM, so 625 μM glutamate was utilized in the final assay to ensure saturation for the PSAT1/PSPH coupling reaction (Figure 2g). The final, fully-coupled PHGDH assay demonstrated robust, linear turnover over 20 minutes, suggesting it was amenable to screening ( Figure 2h). These conditions represent the final coupled PHGDH assay utilized to conduct all subsequent screening and validation work (Table 1).

Pilot Screening
To compare the performance of the optimized assay to the original uncoupled version, preliminary screening was conducted using the LOPAC ®1280 library. In this pilot screen, the uncoupled NADH assay resulted in interference from significant compound fluorescence, which was of a considerably greater magnitude than the NADH fluorescence resulting from PHGDH turnover and NADH production ( Figure 3a). Using ∆ 30 min data at 57 µM compound, where inhibition ≥ 50% was defined as active, the NADH-fluorescence assay yielded 103 hits out of 1280 compounds (hit rate of 8%). In contrast, the PSAT1/PSPH/diaphorase-coupled assay showed nearly 3-fold fewer autofluorescent compound artifacts, (37 out of 1280 compounds, hit rate of 3%) and at a comparatively lower magnitude relative to resorufin signal.
During these screens, plates were measured every 10-15 min for a total of 30 min, and 32 positive control (10 nM PHGDH enzyme) and negative control (no PHGDH enzyme) wells were included on each 1536-well plate to monitor assay performance. The uncoupled assay (direct

Primary Quantitative High-throughput Screening
The coupled PHGDH assay was utilized to screen our in-house collection of more than 400,000 small molecules. In this qHTS screen, the inhibition associated with each well was computed from the endpoint and normalized against control wells. The percent inhibition at each of the concentrations of inhibitor tested was fit to a sigmoidal dose-response curve (Hill equation) using in-house-developed software (https://tripod.nih.gov/curvefit/) to determine the compound IC 50 values. Across >500 1536-well plates, assay performance was strong with Z' generally >0.7 and S:B >20 (Figure 4a). Analysis of these dose-response curves resulted in 1342 high-quality actives that showed strong inhibition with curve classes 1-3, efficacy over 50% and IC 50 ≤ 20 µM ( Figure 4b). These hit compounds were then thoroughly evaluated for both promiscuity and synthetic tractability to eliminate compounds that did not offer good starting points for lead optimization. A total of 239 prioritized inhibitors were reacquired and screened in 11-point doseresponse in the primary screening assay to confirm their activity.

Hit Validation and Triage
A number of secondary assays and counterscreens were selected and utilized to refine and triage initial screening hits (assay hit triage funnel shown in Figure 4c). This triage panel included 1) validation in the primary assay using expanded 11-point dose-response, 2) triage of PAINS [16][17] or other chemotypes/scaffolds not amenable to medicinal chemistry optimization, 3) orthogonal validation in the original uncoupled PHGDH assay (PHGDH alone, using NADH fluorescence), 4) counterscreens against diaphorase alone, 5) orthogonal screening against the coupled PHGDH/PSAT1/PSPH reaction using NADH fluorescence (without diaphorase), and 6) The activity of 1 was subsequently characterized in dose-response using the outlined assay triage hit funnel ( Figure 4d). The primary screen and secondary cherry-pick validation activity of 1 was found to be in close agreement (14.1 μM vs 15.3 μM). In the uncoupled PHGDH assay, which utilized diaphorase and resazurin for detection without PSAT1 and PSPH, 1 displayed an increase in potency (5.4 μM), suggesting that its activity was independent of the presence/activity of the downstream enzymes. Similarly, 1 retained activity in the orthogonal PHGDH assay (17.2 μM), which included PSAT1 and PSPH but utilized NADH fluorescence as a direct measure of turnover, demonstrating that its activity was not dependent on the diaphorase/resazurin coupling system. This observation was further confirmed in the diaphorase counterscreen assay, which showed no appreciable inhibition of diaphorase (<9.3% inhibition at the top concentration of 57.5 μM). Taken together, the assay panel suggested that 1's activity was directly dependent on PHGDH, rather than any of the coupling enzymes.
Select hits were advanced for selectivity testing, which included a small panel of dehydrogenases (namely IDH1, LDHA and GAPDH) to exclude pan-dehydrogenase inhibitors. A number of hit compounds were found to demonstrate strong selectivity toward PHGDH over other dehydrogenases, with little comparative off-target activity. Notably, the screening hit 1 displayed complete selectivity toward PHGDH, with no discernable off-target dehydrogenase inhibition ( Figure 4e).

Medicinal Chemistry & SAR Exploration
Based on this panel of assays, a series of 2-pyridinyl-N-(4-aryl)piperazine-1-carbothioamides emerged as the top, validated chemotype, as low micromolar inhibitors of PHGDH. This chemotype was explored earlier during another NCGC structure-activity campaign to develop inhibitors of bacterial phosphopantetheinyl transferase (PPTase) 18 . Utilizing compounds from this previous medicinal chemistry effort, we were able to rapidly establish preliminary structureactivity relationships (SAR) beyond hit 1 ( Table 2).
We began our PHGDH SAR investigations with a focus on the necessity of the thiourea group (Table 2). This feature of the pharmacophore proved essential to activity as the urea (3), guanidine (4), and thioamide (5) replacements led to significant loss of PHGDH activity.
Additionally, other minor modifications to this region of the molecule were not tolerated such as N-methylation (2), shortening of the linkage from the thiocarbonyl to the pyridine (6), or lengthening the linkage (7).
Subsequently, we shifted our SAR analysis to the aminopyridine portion of the chemotype (Table 3). The pyridine in this region was required to maintain target inhibition as phenyl variant 8 lost all potency. The positioning of the pyridine nitrogen also proved to be optimal at the two-position as the 3-and 4-pyridyl analogs 10 and 11 lost both potency and efficacy in comparison to compound 9. A second nitrogen proximal to the thiourea nitrogen in the form of a pyrimidine (12) led similarly to diminished PHGDH activity as did substitution of the pyridine with electron-withdrawing groups (analogs 13 and 14). The addition of a second methyl group resulted in an improvement in potency, whereas expansion of the pyridine ring system in the form of a quinoline (analogs 16 and 17) showed small perturbations in target activity.
A key feature of our approach to the optimization of this chemotype involved the evaluation of structure-property relationships (SPR) in parallel to all of our SAR work. We performed a set of in vitro absorption, distribution, metabolism and excretion (ADME) assays on all compounds including aqueous kinetic solubility, rat liver microsomal stability (single point), and parallel artificial membrane permeability assay (PAMPA). At this stage of our investigations, we leveraged improved SPR observations as compound 15 displayed improved microsomal stability (rat liver microsomes t 1/2 = >30 min) compared to analog 9 (t 1/2 = 16 min). Thus, dimethylpyridine was carried forward for subsequent SAR studies on the remaining regions of the chemotype. Furthermore, although only a small sample of analogs probing this section of the hit molecule 1 are displayed here, a broader exploration carried out later in the SAR campaign confirmed, in the same way, that this chemotype would not accommodate alterations to the pyridyl ring system.
Having established that the 2-pyridine and the thiourea are important features of the pharmacophore, we shifted our attention to the arene affixed to the piperazine in order to assess the flexibility therein ( Table 4). Movement of the trifluoromethyl group around the arene proved to be well-tolerated with the best location being para to the piperazine (19). Unsubstituted With insufficient improvements in target potency and in vitro ADME properties through early SAR and SPR studies, we considered more significant modifications to the piperazine core in conjunction with changes to the group bridging to the trifluoromethyl phenyl ring. Along these lines, we installed a one-carbon spacer (both methylene and carbonyl) from the piperazine core to the arene segment. Additionally, we excised the piperazine nitrogen distal to the thiourea to a pair of amino piperidines (both 3-and 4-substituted). Methods to access many of these differentially substituted cores were reported previously. 8,18 The majority of the piperazine analogs maintained much of their activity (15, 19, 30, 31) with significant drop-offs being observed only with amides 28 and 29. The optimal 3-amino piperidine analog was that of the p- To further probe the ethylene diamine feature of the core, a subset of ring-opened analogs were evaluated. The increased flexibility resulted in a loss of activity with the dimethylated analog 44 maintaining moderate potency (Table 6). Conversely, rigid variations of the core, such as the fused bis-pyrrolidine analog 45 and 1,4-diazepane 46, were also less active than NCT-503 (31).
Returning to the piperazine and the methylene linkage to the trifluoromethyl arene as above, we wanted to investigate whether substitution (47) Table 7). Testing of this large set of analogs focused on this aryl region of the chemotype proved unable to improve potency.

Discussion
The assay design and optimization described here outlines a novel coupled PHGDH assay with robust performance and improved enzymatic turnover. This protocol was developed to avoid many of the fluorescent artifacts and issues common to classic NADH-linked dehydrogenase detection methods. This novel assay enabled a large-scale ultra high-throughput screen that yielded validated small molecule PHGDH inhibitors, including one particular chemotype of interest. The systematic structural exploration and optimization of this class of 2-pyridinyl-N-(4aryl)piperazine-1-carbothioamides with parallel SAR and SPR campaigns working in concert led to the discovery of NCT-503 (31) as a selective probe for the detailed evaluation of human PHGDH biology both in vitro and in vivo. Emerging from a structural class originally discovered and optimized for the inhibition of bacterial phosphopantetheinyl transferase (PPTase), this scaffold was selectively repurposed and optimized for PHGDH activity. Along these lines, a panel of small molecules synthesized as part of PPTase SAR campaign was tested against both enzymes, and no correlation in activity was observed between the two targets ( Figure 5). In addition to an increase in potency over both the hit molecule and that of NCT-502 (27), moderate improvements in both solubility and microsomal stability were also observed for NCT-503 (31) ( Table 8).
These tool compounds complement an existing set of hPHGDH probes 3 (Figure 1) discovered through a variety of techniques including high-throughput screening 6 , fragment-based lead generation 5 , structure-based drug design 10 , as well as a merger of both fragment screening and pharmacophore convergence 11 . This class of inhibitors has expanded our understanding of the role of PHGDH in tumor metabolism. The uncoupled variation of the PHGDH assay was optimized to run in the absence of PSAT1 and PSPH downstream enzymes.

Experimental Procedures
PHGDH inhibition data was analyzed by calculating delta RFUs (increase in resorufin fluorescence between T = 0 and T = 20 min) and normalizing to the delta RFU values of 0× and 1× PHGDH enzyme controls as 100% and 0% PHGDH inhibition, respectively. 32 wells of each 1536-well assay plate were dedicated to each of these 0× and 1× PHGDH controls. These The pure fractions were combined and most of the organic portion removed in vacuo. To the resulting mixture was added dichloromethane as well as saturated aqueous sodium bicarbonate solution, in order to free base the product. The layers were separated and the aqueous layer was re-extracted with dichloromethane three additional times. The combined organic layers were dried with MgSO 4 and concentrated in vacuo to afford N-(4,6-dimethylpyridin-2-yl)-4-(4-(trifluoromethyl)benzyl)piperazine-1-carbothioamide (NCT-503, 31, 0.809 g, 28% over 3 steps) as a colorless foam. 1   The diaphorase assay was run using a resazurin titration to determine the substrate's Km (0.05 mM), and feedback inhibition was observed at substrate concentrations above 0.25 mM. c) PHGDH activity (10 nM) was measured utilizing varying stoichiometries of PSAT1 and PSPH to relative contributions to PHGDH assay turnover. Conditions for the coupled PHGDH assay were determined by testing titrations of d) diaphorase, e) 3PG, f) NAD + and g) glutamate to identify ideal assay parameters (i.e. Kms for the substrates 3PG and NAD + , and non-limiting concentrations for the coupling reagents diaphorase and glutamate). h) A time course demonstrating robust, near-linear kinetics is shown for the fully-coupled assay in the presence and absence of diaphorase. Figure 3. PHGDH preliminary screening and optimization. a) Plate images are shown for T=0 (left plates) and T=30 min (right plates) reads from two PHGDH screens against the LOPAC library (11.5 μM). The two top plates demonstrate the autofluorescence of LOPAC library compounds in the PHGDH assay utilizing NADH-based detection, while the bottom two plates show well signal from the PHGDH assay using resazurin-based detection. Very few autofluorescent compounds can be observed in the resazurin wavelength, providing a clearer landscape for observing potential PHGDH inhibition. b) Examples are shown of false-positive hits identified from LOPAC in the PHGDH assay utilizing NADH-based detection. All corresponding compounds were completely inactive in the diaphorase/resazurin-based PHGDH assay, demonstrating an advantage of migrating to the 598 nm detection realm. Assay signal windows observed during LOPAC screens using c) NADH-linked or d) resorufin-based readouts.    ViewLux fluorescence read (ex525/em598, 5 sec exposure, 5000 excitation energy) 5 Room temperature