Development of substituted benzimidazoles as inhibitors of Human Aldehyde Dehydrogenase 1A Isoenzymes

Abstract

Aldehyde dehydrogenase 1A (ALDH1A) isoforms may be a useful target for overcoming chemotherapy resistance in high-grade serous ovarian cancer (HGSOC) and other solid tumor cancers. However, as different cancers express different ALDH1A isoforms, isoform selective inhibitors may have a limited therapeutic scope. Furthermore, resistance to an ALDH1A isoform selective inhibitor could arise via induction of expression of other ALDH1A isoforms. As such, we have focused on the development of pan-ALDH1A inhibitors, rather than on ALDH1A isoform selective compounds. Herein, we report the development of a new group of pan-ALDH1A inhibitors to assess whether broad spectrum ALDH1A inhibition is an effective adjunct to chemotherapy in HGSOC. Optimization of the CM10 scaffold, aided by ALDH1A1 crystal structures, led to improved biochemical potencies, improved cellular efficacy as demonstrated by reduction in ALDEFLUOR signal in HGSOC cells, and substantial improvements in liver microsomal stability. Based on this work we identified two compounds 17 and 25 suitable for future in vivo proof of concept experiments.

Graphical Abstract

1. Introduction

Ovarian cancer is the 5^th most deadly cancer for females in the US, and worldwide accounted for more than 200,000 deaths in 2020 [1]. Despite the development and approval of new, targeted therapies for this disease, there has been only marginal overall improvement in clinical outcomes relative to the long-standing standard of care: surgical debulking of the tumor, combined with platinum- and taxane- based chemotherapy [2–6]. In particular, inhibitors of poly-ADP-ribosylation have improved clinical outcomes for patients with Breast Cancer gene (BRCA) mutations [7]. However, the majority of ovarian cancer patients do not harbor tumorigenic BRCA mutations [8], so their standard of care remains surgery and chemotherapy. With combined surgery and chemotherapy, while most patients will have a complete clinical response, most patients will relapse and ultimately develop chemotherapy resistant disease. Recurrence presumably arises from a small population of resistant tumor cells that survive the initial treatment regimen. Consequently, the ability to augment the effectiveness of the initial chemotherapy to eliminate resistant cells could greatly improve the outcomes of patients with ovarian cancer [9].

While it is clear there are multiple mechanisms through which resistance can arise, one potential clinical target to overcome chemotherapy resistance is the aldehyde dehydrogenase 1A family of enzymes (ALDH1A). ALDH1A enzymes can directly metabolize and inactivate chemotherapeutics such as cyclophosphamide and are known to directly metabolize cellular products of oxidative stress/damage [10–12]. ALDH1A enzymes are also linked to both the cellular stress response and DNA damage repair such that they are also likely linked with resistance to other chemotherapies such as platinums and taxanes [13, 14]. ALDH expression has even been linked with resistance to therapeutics such as tyrosine kinase inhibitors [15].

The apparent linkage between ALDH1A enzyme expression and chemotherapy resistance may relate to the presence of ALDH1A enzymes in cancer stem-like cells (CSCs). The ALDEFLUOR assay has been used to identify CSCs in ovarian and other solid tumor cancers [16–20]. This assay directly measures ALDH enzyme activity in cells when the ALDEFLUOR reagent is oxidized by an ALDH isoenzyme to the acid form which is less permeable to membranes and accumulates within the target cell population. The ALDEFLUOR reagent can be metabolized by several ALDH isoenzymes [21, 22], with the ALDH1A isoenzymes of specific interest for cancer stem cells. We and others have postulated that ALDH plays a functional role in CSC biology and demonstrated that small molecule inhibitors of ALDH in combination with chemotherapy may offer an avenue to target CSCs as a means to improve patient outcomes [23–26]. The three ALDH1A isoforms, ALDH1A1 (1A1), ALDH1A2 (1A2), and ALDH1A3 (1A3), exhibit a high degree of sequence identity (>70%), and are of particular interest in targeting CSCs. Both the 1A1 and 1A3 isoenzymes are linked to stem-like characteristics in cancers arising from a variety of tissue types and have been implicated in chemoresistance [16, 17, 26–32]. High-grade serous ovarian cancer comprises the histologic subtype responsible for 70–80% of ovarian cancer deaths and demonstrates high levels of 1A3 expression, as well as a modest elevation of 1A1 [33, 34].

Given the potential for combination therapeutic approaches and the need to better understand ALDH biology, the discovery and development of ALDH inhibitors has expanded over the past decade. The landscape surrounding these efforts has been comprehensively reviewed [10, 35, 36]. Notably, a concerted medicinal chemistry effort to develop ALDH1A1 selective inhibitors culminated in the discovery of NCT-506 and several analogs [37, 38] (Fig. 1). This orally bioavailable compound potently inhibits ALDH in the 1A1-expressing OV-90 ovarian cancer cell line (ALDEFLUOR IC₅₀ = 161 nM) and reverses taxane resistance in a SKOV-3 resistant cell line. Although originally discovered as inhibitors of ethanol metabolism and mitochondrial ALDH2, modifications of the daidzin scaffold has produced a number of moderately potent and very selective inhibitors of ALDH1A3 [39, 40], analogs of which have shown efficacy in a glioblastoma mouse model [41, 42].

Fig. 1.

The structures and inhibition characteristics of selected ALDH1A inhibitors. The inhibition data are summarized from the literature for NCT-506 [37], compound 69 [43] and CM10 [23].

The overarching goal of our research program has been to explore the structure-activity relationships of ALDH1A isoform selectivity using different starting scaffolds. Given the role(s) of individual ALDH1A isoforms in chemoresistance and the implication that multiple ALDH1A isoforms are identified in ovarian cancer as well as the elevation of 1A3 in high-grade serous tumors, we have recently sought to develop selective as well as more broad-spectrum ALDH1A inhibitors (e.g. 69, Fig. 1) to facilitate the study of ALDH1A biology in a wider array of cell lines [24, 43]. In this report we disclose our efforts toward the development of viable in vivo probes from a new chemical series by improving in vitro potency for ALDH1A isoenzymes, improving ALDH1A target engagement in cell lines by monitoring ALDEFLUOR inhibition, and optimizing metabolic stability of an initial high throughput hit compound (CM10, 1, Fig. 1 [23, 44]). These characteristics were assessed through measuring biochemical inhibition of the three ALDH1A isoforms and via inhibition of the ALDEFLUOR signal in the high-grade serous ovarian cancer cell line PEO1 that expresses high levels of ALDH1A3 [23]. To monitor selectivity, we also tested compounds against the most homologous ALDH isoform not included in the ALDH1A family, ALDH2, and none of the compounds in this study inhibited ALDH2 by more than 40% of control activity at 20 µM. Metabolic stability (mouse liver microsomes) was assessed for key compounds to monitor progress toward suitable compounds for future in vivo studies.

2. Materials and methods

2.1. Compound synthesis and characterization

Data to support new chemical synthesis and characterization of new compounds can be found in the Supplemental Methods.

2.2. Protein purification and enzymatic assays

Human ALDH1A1, ALDH1A2, and ALDH1A3 were prepared and purified as previously described [45–49]. Inhibition of ALDH activity by compounds and EC₅₀ curves were determined by measuring the formation of NAD(P)H spectrophotometrically at 340 nm (molar extinction coefficient of 6200 M^-1 cm^-1) on a Cary 50 spectrophotometer and on a Spectramax 340 PC plate-reader using purified recombinant enzyme. Reaction components for assays consisted of 100–200 nM enzyme, 200 µM NAD⁺, 100 µM propionaldehyde, and 2% DMSO in 25 mM BES buffer, pH 7.5. All assays were performed at 25 °C and were initiated by addition of substrate after a standard two-minute incubation period. This incubation period was varied for compounds that possess potentially reactive groups (compound 1, 17 and 20). EC₅₀ curves were collected for compounds which substantially inhibited ALDH1A activity at 20 µM compound. Data were fit to the four parameter EC₅₀ equation $\left[\text{y}={\text{y}}_{\text{min}}+({\text{y}}_{\text{max}}-{\text{y}}_{\text{min}})/\left(1+10^{\left([\text{I}]-{\text{logEC}}_{50}\right)}^{\text{HillSlope}}\right)\right]$ using SigmaPlot (v14), and the values represent the average/SD of three independent experiments (each experiment having n = 3 observations per dose). Based on these fits, the average maximal extent of inhibition for all compounds evaluated via dose-responses were 77 ± 10% for ALDH1A1 (Supplemental Table 1), 90 ± 10% for ALDH1A2 and 90 ± 10% for ALDH1A3. To evaluate the mechanism of inhibition (competitive, non-competitive, uncompetitive), co-variation experiments were performed in which the concentration of the inhibitor was varied against varying concentrations of one of the co-substrates (either NAD⁺ or aldehyde substrate) at fixed concentrations of the non-varied co- substrate. NAD⁺ was varied at fixed acetaldehyde (750 µM) acetaldehyde was varied at fixed NAD⁺ (500 µM). Acetaldehyde, rather than propionaldehyde, was used as the aldehyde substrate for all covariation experiments as the values for acetaldehyde are higher and do not approach the limits necessary to account for tight-binding kinetics. Each individual experiment was designed to evaluate the impact of 4 different inhibitor concentrations and 4 to 5 concentrations of the varied substrate with each set of concentrations measured in triplicate (48 to 60 independent measurements per experiment). Each experiment was performed three times. The data from each experiment was fit globally by non-linear regression and evaluated for goodness of fit to the complete velocity equations for competitive, non-competitive, mixed- type inhibition and uncompetitive inhibition, as well as the same equations modified for situations where the inhibition is not complete (partial inhibition). Decisions on which mechanism best described the observed data were based on R² values for fits to all 8 equations from each of the three complete experiments and an evaluation of whether clear systematic deviations were apparent in the graphical fits to those equations, and in cases where mixed inhibition was the best fit, an evaluation of whether alpha*K_i was within the range of concentrations tested in the experiment; if not, a simpler model (competitive, non-competitive or uncompetitive was chosen). The reported K_i values are the averages of the best mechanism globally fit across the three complete experiments.

2.3. Crystal structure determination

The structure of ALDH1A1 bound to compound 4 was determined using crystals of the naturally occurring polymorphic variant, N121S. Crystals of this variant are frequently more amenable to capturing inhibitors bound within the active site, but the mutation itself has little to no impact on the kinetic or global structural characteristics of the enzyme [44]. The crystals were grown by equilibrating 4–8 mg/mL ALDH1A1 N121S or wild-type enzyme against 100 mM sodium BisTris, pH 6.2–6.5, 6–11% PEG3350, 200 mM NaCl, and 5–10 mM YbCl₃. The crystal of ALDH1A1 N121S in complex with compound 4 (Table 1) was prepared by co-crystallizing in the presence of 500 µM 4 (2% v/v DMSO final) and then incubating the co-crystal in the presence of 1 mM NAD⁺ for 1 hour prior to flash-freezing in 20% (v/v) ethylene glycol in ligand solution. The crystal of wild-type ALDH1A1 in complex with NCT-506 (Table 1) was prepared by co-crystallizing in the presence of 500 µM of compound (2% v/v DMSO final) and flash-freezing in 20% (v/v) ethylene glycol added to the crystallization solution. All diffraction data were collected at Beamline 19-BM operated by the Structural Biology Consortium at the Advance Photon Source (APS), Argonne National Laboratory. Diffraction data were indexed, integrated, and scaled using HKL3000 [50]. The PHENIX program suite (v1.17) was used for refinement and validation [51–53]. The Coot (v0.8.9.2) molecular graphic application was used for model building [54].

Table 1.

Optimization of the substituted phenol group.

				ALDH EC50 (nM) or % Control at 20 µM			CLogP	PEO1 Cell ALDEFLUOR (% inhibition)
No.	R1	R2	X1	1A1	1A2	1A3
1	−OH	−CH2CH=CH2	NH	1700 ± 240	740 ± 210	640 ± 160	5.04	ND
2	−OH	−H	NH	3700 ± 380	2700 ± 320	700 ± 45	3.94	ND
3	−H	−OH	NH	73%	85%	87%	3.94	ND
4	−OH	−OCH3	NH	750 ± 74	9500 ± 890	620 ± 67	3.78	ND
5	−OH	−F	NH	3000 ± 50	1000 ± 160	1300 ± 620	3.86	ND
6	−OH	−Cl	NH	190 ± 15	240 ± 57	79 ± 30	4.27	50 ± 4% (10 µM)
7	−OH	−Cl	O	76%	56%	110%	4.91	ND

2.4. ALDEFLUOR Assay

PEO1 cells were grown and ALDEFLUOR (STEMCELL Technologies) assays performed as previously described [24]. Briefly, cells were cultured to 70–80%, trypsinized, washed with PBS, and then resuspended in ALDEFLUOR buffer. ALDEFLUOR reagent was added, cells were rapidly mixed and then equally distributed into tubes containing inhibitor, DEAB control, or vehicle. After a 30 min incubation at 37 °C, cells were washed with ALDEFLUOR buffer and maintained on ice until flow- cytometric analysis. Gating was based on DEAB (inhibitor control-set to <1%) and vehicle control treated cells (positive control). The percent of ALDH inhibition was set as percentage of ALDEFLUOR positive cells for a particular sample and the percentage of ALDEFLUOR positive cells versus vehicle treated control. All values are expressed the average of three independent experiments with the corresponding standard deviation.

2.5. Metabolic stability in mouse liver microsomes

The metabolic stability was assessed using CD-1 mouse liver microsomes (MLM). 1 µM of each compound was incubated with 0.5 mg/mL microsomes and 1.7 mM cofactor β-NADPH in 0.1 M phosphate buffer, pH 7.4, containing 3.3 mM MgCl₂ at 37 °C. The DMSO concentration was less than 0.1% in the final incubation system. At 0, 5, 10, 15, 30, 45, and 60 min of incubation, 40 µL of reaction mixture were taken out, and the reaction was quenched by adding 3-fold excess of cold acetonitrile containing 100 ng/mL of internal standard for quantification. The collected fractions were centrifuged at 15000 rpm for 10 min to collect the supernatant for LC−MS/MS analysis, from which the amount of compound remaining was determined. The LC−MS analysis was performed by a Shimadzu HPLC system equipped with a Waters XBridge-C18 column (5 cm × 2.1 mm, 3.5 µm) as the front-end to an AB Sciex QTrap Model 5500 mass spectrometer equipped with an electrospray ionization source (Applied Biosystems, Toronto, Canada) operated in the positive-ion multiple reaction monitoring (MRM) mode for detection. The natural log of the amount of compound remaining was plotted against time to determine the disappearance rate and the half-life of tested compounds.

3. Results and discussion

Previous work in our laboratories had identified compound 1 (Fig. 1, Table 1) as a lead inhibitor from a high-throughput screen [44] that inhibited ALDH1A1, ALDH1A2 and ALDH1A3 in vitro and selectively depleted CD133⁺ A2780 cells in culture [23]. In this study we sought to further develop this series by exploring the potential of the benzimidazole scaffold to support the development of both pan-ALDH1A and isoenzyme selective inhibitors and improve their cellular efficacy toward the target enzymes, as well as their metabolic stability to enable in vivo proof of concept studies.

3.1. Chemistry

New compounds were prepared in a regioselective manner as summarized in Scheme 1. Commercial 2- nitroanilines A-1 were N-alkylated via the corresponding N-trifluoroacetamides A-2 under basic conditions, affording N-alkyl-2-nitroanilines A-3 [55]. Reduction of the nitro group followed by cyclization with cyanogen bromide resulted in formation of 2-aminobenzimidazoles A-6. Conversion to final analogs was completed by reductive alkylation of the free amine with various benzaldehydes. Key intermediates A-3 could alternatively be obtained directly from 2-fluoronitrobenzenes A-4 by heating with amines R⁶NH₂. Intermediates A-6 could also be obtained by direct N-alkylation of aminobenzimidazoles A-7, but this was not regioselective with respect to aromatic substituents on A-7, requiring difficult chromatographic separation of regio-isomers. This approach was therefore limited to just N-alkyl analogs derived from simple unsubstituted 2-aminobenzimidazole. Installation of an ethynyl group at R⁸ required the preparation of intermediate B-2 from commercial aldehyde B-1 (Scheme 2, [56]. Conversion of B-2 to 20 was then completed in the same manner as conversion of A-3 to final analogs in Scheme 1. Finally, ether analog 7 could be prepared by direct alcohol displacement of chloride from commercially available 2-chlorobenzimidazole C-1 (Scheme 3) under strongly basic conditions.

Scheme 1.

Reagents and Conditions: a) TFAA, CH₂Cl₂, sat’d NaHCO₃, 0°C – RT; b) Bu₄NBr, R⁶Br, toluene, aq NaOH, 80°C; c) R⁶NH₂, Cs₂CO₃, CH₃CN, 70 °C; d) Zn powder, MeOH, AcOH, RT; e) BrCN, EtOH, 60°C; f) KOH, 2:1 THF:EtOH, R⁶Br, 65°C, 24h; g) ArCHO, toluene, reflux with azeotrope; h) NaBH4, MeOH, 0°C – RT.

Scheme 2.

Reagents and Conditions: a) CH₃COC(=N₂)PO(OMe)₂, K₂CO₃, MeOH, RT.

Scheme 3.

Reagents and Conditions: a) 2-chloro-6-(hydroxymethyl)phenol, NaH, THF, 50°C.

3.2. Optimization of the substituted phenol group

CM10 (compound 1), was reported as a partial inhibitor of ALDH1A1 [23, 44], as are many of the new compounds reported in this manuscript. In contrast, those same compounds inhibit the 1A2 and 1A3 isoenzymes, on average, 90% or more. We report the average level of maximal inhibition for compounds where dose responses were obtained for ALDH1A1 in Supplemental Table 1. We first examined the requirements at the R¹ and R² positions and the bridging heteroatom between the benzimidazole and phenol rings for enzymatic potency. A hydroxyl at R¹ is required and cannot be replaced by a hydroxyl at the R² position (Table 1, compound 2 versus 3). We next examined whether the vinyl group at the R² position could be varied. Removal of the bulky vinyl group (2) was detrimental to potency for 1A1 and 1A2 but did not significantly impact potency toward 1A3. The optimal substituent at R² is a chlorine (e.g. 6) which improves the potency toward ALDH1A1 and 1A3 by 8-fold and by approximately 3-fold for 1A2. The chloro substituent is likely the right combination for enhancing hydrophobicity of the ring and acidity of the hydroxyl and optimizing interactions with the region surrounding position 2. This compound was able to reduce the ALDEFLUOR signal in PEO1 cells by 50% at 10 µM. A fluorine substituent (5) may be too polar for the site or perhaps too strongly electron withdrawing, thereby rendering the pK_a of the phenol hydroxyl too low for the relatively hydrophobic characteristics of the ALDH1A substrate binding site. Replacement of the amine linking group by an ether linkage (7) destroyed potency for all 1A isoenzymes, possibly by an inductive effect that lowers the hydrogen- bonding capacity of the imidazole nitrogen.

3.3. Structure of compound 4 bound to human ALDH1A1

To better correlate the structure-activity relationships of this series with the structural features of the target enzymes, we solved the X-ray crystal structure of compound 4 bound to N121S ALDH1A1 in a complex with coenzyme (Table 2). The bound coenzyme is held in the non-productive ‘hydrolysis position’ that is common in ALDH structures [57]. Compound 4 is bound within the substrate-binding site with the benzimidazole ring oriented toward the active site nucleophile, C303, in the inner region of the substrate-binding site (Fig. 2A). The benzimidazole is sandwiched between V460, I304 and the aromatic side chains of F466, Y297 and F171 (the latter two residues can be seen in Fig. 2B and 2C, respectively). The substituted phenol group is positioned such that the hydroxyl and -OMe groups are oriented toward the indole nitrogen of W178 and are in position to accept a hydrogen bond from the indole nitrogen (3.4 Å and 3.1 Å, respectively). The side chain hydroxyl of T129 is in position to donate a hydrogen bond to the -OMe group (3.1 Å), though this hydrogen bond must not be critical, since a chloro- substituent at this position improves potency (Table 1, compound 4 versus 6). The aromatic face of the phenol ring forms van der Waals contacts with G125 (Fig. 2A and 2B). A critical aspect of the bound conformation of compound 4 is that the phenol hydroxyl is in position to donate an internal hydrogen bond to the benzimidazole nitrogen (2.8 Å, Fig. 2A). Given that the hydroxyl group is required and that non-polar groups are tolerated at the 2-position (Table 1), it is likely that this internal hydrogen bond explains the absolute requirement for the hydroxyl group and that this internal hydrogen bond makes the phenolic group a better hydrogen bond acceptor from the indole nitrogen of W178 (Fig. 2A). The N- propyl alkyl chain attached to the benzamidazole group extends into the pocket bordered by G458 in 1A1, but unlike compounds selective for 1A1 [24, 37, 43, 49] does not fully occupy the site. To explore the relationship between compounds that fully occupy the site bordered by G458 and isoenzyme selectivity, we also determined the crystal structure of NCT-506 [37] bound within wild-type human ALDH1A1 to 1.75 Å resolution (Table 2). Like the theophylline-based scaffolds [38, 49], the quinoline ring structure is bound within the pocket defined by F171, Y297, I304, V460 and G458 and fully occupies this site (Fig. 2C). In contrast, the benzimidazole scaffold in this report, and the pyrazolo-pyrimidine scaffold in our prior work [24, 43], possess tunable extensions to the central scaffold that can be optimized for different levels of selectivity toward ALDH1A isoenzymes. Another striking aspect of NCT-506 binding is the engagement of its nitrile group via hydrogen bonding interactions with the main chain amide nitrogen atoms of residues C303 and I304 within the active site loop. However, there is no evidence of direct engagement with the catalytic nucleophile, itself (C303).

Table 2.

Crystallographic Data Statistics.

	Compound 4	NCT-506
PDB Code	8T0N	8T0T
Data Collection (ADSC Quantum Q210)
Date of Collection (APS Beamline 19-BM; λ = 0.98 Å)	08/2016 19-BM	11/2018 19-BM
Space Group	P422	P422
Cell Dimensions
a,b,c (Å)	109,109,83	109,109,83
α,β,γ (deg)	90,90,90	90,90,90
Resolution (Å)	37.0–1.86	35–1.75
Rmerge	0.087 (0.48)	0.071 (0.65)
Rmeas	0.094 (0.53)	0.075 (0.71)
Rpim	0.035 (0.22)	0.025 (0.26)
CC1/2 (highest shell)	0.865	0.787
I/σ<I>	17.2 (3.2)	24.8 (3.0)
Completeness (%)	99.5 (96.7)	99 (99)
Redundancy	6.8 (5.5)	9.2 (7.0)
Refinement
Number of reflections (Rw)	42,676	51,103
Number of reflections (Rf)	1999	2672
Number of protein atoms	3891	4134
Number of water molecules	247	259
Number of inhibitor molecules	1	1
Occupancy of inhibitor(s)	1.0	1.0
Rwork/Rfree	17.5/22.2	19.8/24.0
R.M.S.D. from Ideal Values
Bond Length (Å)	0.012	0.007
Bond Angle (deg)	1.18	0.94
Ramachandran plot
Preferred (%)	97.2	97.0
Outliers (%)	0.2	0.2
Clashscore	3.8	3.1
Average B (Å2)	34.8	31.8
Protein	34.7	31.8
Inhibitor	61.9	25.7
Solvent	32.9	32.1

Fig. 2.

Structure of ALDH1A1 with bound ligands. (A) Structure of 4 bound to human ALDH1A1. The electron density from the original m2Fo-Fc map (contoured at 1 standard deviation of the map) prior to the inclusion of 4 in the structure factor calculation. (B) A surface rendering of ALDH1A1 with 4 (cyan sticks) bound within the substrate binding pocket. (C) Structure of NCT-506 bound to human ALDH1A1. The electron density from the original m2Fo- Fc map (contoured at 1 standard deviation of the map) prior to the inclusion of NCT-506 in the structure factor calculation. (D) A structural overlay of the binding modes of 4 and NCT-506. Dashed black lines indicate potential hydrogen bonding interactions, Panels A and C.

3.4. Optimization of aryl substituents

Having established optimal substituents for the R¹ and R² positions and noting that there is additional space for binding interactions available on the solvent exposed side of the phenol group, we next examined whether substitutions to the phenol ring at the opposing R⁴ and R⁵ positions could improve potency (Table 3). Despite the available space within the binding pocket, addition of individual substituents to the unsubstituted phenol 2 at positions R⁴ (compounds 9–12) was detrimental to potency for the 1A1 and 1A2 isoenzymes, although some substituents at the R⁴ position were better tolerated by ALDH1A3. This behavior was distinct from the addition of a fused ring at positions 4 and 5 (compound 8), which improved potency for 1A1 relative to compound 2, but was not as favorable for 1A2 or 1A3. Similarly, adding halogen substituents at position R⁴ to the chlorophenol 6 did not yield any significant improvement in potency (compounds 13 and 14). It is possible that addition of another halogen to the 2- chlorophenol ring lowers the pK_a of the phenolic hydroxyl sufficiently to eliminate or reduce the strength of the intramolecular hydrogen bond with the benzimidazole nitrogen or impacts the interaction with the indole nitrogen of W178 in the binding site (Fig. 2A). Compound 8 from this group was further characterized by steady state inhibition kinetic analyses versus both varied acetaldehyde substrate at fixed NAD⁺ and versus varied NAD⁺ at fixed acetaldehyde. Compound 8 was found to be a partial non- competitive inhibitor versus acetaldehyde and an partial uncompetitive inhibitor versus NAD⁺ with average K_i values (n = three experiments) of 690 ± 110 nM and 440 ± 50 nM, respectively, and an average maximal inhition across the 6 experiments of 86% (beta = 0.14 ± 0.04) (Fig. 3). The uncompetitive pattern toward NAD⁺ is consistent with compound 8 binding to a different enzyme species in the catalytic cycle than does NAD⁺, which is consistent with the binding of compound 8 in the aldehyde substrate binding site (Fig. 2). The non-competitive inhibition pattern of 8 toward aldehyde is more difficult to explain, but might result from binding to an enzyme species where NAD⁺ is not productively bound to facilitate hydride transfer, as observed for the structure with compound 4. The reaction mechanism involves NAD⁺ binding to the enzyme and the coenzyme undergoing a conformational change prior to assuming the productive conformation required for hydride transfer [45, 48, 57–60]. It is possible that 8 binds to one of the non-productively positioned complexes with NAD⁺. Non-competitive inhibition with respect to varied aldehyde concentrations have been observed for multiple chemical scaffolds [45, 49], though the precise mechanistic reason for this remains elusive.

Table 3.

Optimized Aryl Substituents.

				ALDH EC50 (nM) or % Control at 20 µM			CLogP	PEO1 Cell ALDEFLUOR (% inhibition)
No.	R2	R4	R5	1A1	1A2	1A3
2	−H	−H	−H	3700 ± 380	2700 ± 320	700 ± 45	3.94	ND
8	−H			780 ± 78	2200 ± 200	1800 ± 500	5.28	ND
9	−H	−F	−H	2600 ± 400	63%	840 ± 75	4.08	ND
10	−H	−OCH3	−H	36%	47%	41%	3.78	ND
11	−H	−CH3	−H	44%	74%	38%	4.75	ND
12	−H	−Cl	−H	93%	69%	30%	4.80	ND
6	−Cl	−H	−H	190 ± 15	240 ± 57	79 ± 30	4.27	50 ± 4% (10 µM)
13	−Cl	−F	−H	52%	42%	140 ± 21	4.36	59 ± 9% (10 µM)
14	−Cl	−Cl	−H	300 ± 40	3200 ± 1200	520 ± 40	4.73	ND

Fig. 3.

Mechanism of inhibition analysis for compound 8. Representative Lineweaver-Burk plots generated by globally fitting the kinetic data to the complete velocity equation in the absence of products for the inhibition of compound 8 toward ALDH1A1 under conditions of varied NAD⁺ at fixed acetaldehyde (750 µM) (A) and under conditions of varied acetaldehyde at fixed NAD⁺ (500 µM) (B). The best fits for each sets of experiments were partial uncompetitive inhibition against varied NAD⁺ $\left[\text{v}={\text{V}}_{\text{max}}\text{*}\left(1+\text{beta}*\left({\text{I/K}}_{\text{i}}\right)\right)/\left(1+{\text{I/K}}_{\text{i}}+{\text{K}}_{\text{M}}\text{/S}\right)\right]$ with estimated kinetic constants, ${\text{K}}_{\text{M}}=25.1\text{\hspace{0.33em}µM}$, ${\text{V}}_{\text{max}}=6.82\text{ U/mg}$, $\text{beta}=0.104$, ${\text{K}}_{\text{i}}=448\text{ nM}$, and partial non-competitive inhibition against varied acetaldehyde $\left[\text{v}={\text{V}}_{\text{max}}/\left(\left(1+{\text{K}}_{\text{M}}\text{/S}\right)*\left(1+{\text{I/K}}_{\text{i}}\right)/\left(1+{\text{I*beta/K}}_{\text{i}}\right)\right)\right]$ with estimated kinetic constants, ${\text{K}}_{\text{M}}=66.2\text{ µM}$, ${\text{V}}_{\text{max}}=6.10\text{U/mg}$, $\text{beta}=0.141$, ${\text{K}}_{\text{i}}=819\text{ nM}$.

3.5. Optimization of the benzimidazole ring structure

Having optimized the phenol substituent of this series we next examined substitutions to the benzimidazole ring structure. Inspection of the structure of 4 bound to ALDH1A1 (Fig. 2A) reveals that it might be possible to optimize substituents at the R⁷ and R⁸ positions of the benzimidazole ring. This region of the ring system is oriented toward a conserved region of the active site that includes the catalytic nucleophile, C303, and the side chains of E269 and T245 that are generally believed to be responsible for positioning and activating the nucleophilic water that hydrolyzes the acyl-enzyme intermediate [61]. Variation at the R⁸ position produced an interesting pattern with the three 1A isoenzymes. ALDH1A3 was relatively insensitive to the addition of either hydrophobic or polar substituents. In contrast, all substituents examined improved the potency toward ALDH1A2 (Table 4), with the alkynyl and nitrile analogs (compounds 17 and 20) being the most potent toward 1A2. In contrast, ALDH1A1 was more sensitive to the chemical nature of the substituent, with a strong negative selection against polar substituents (compounds 16 or 17 versus 19). Regardless, cellular efficacy was improved by all substitutions at the R⁸ position, with compounds 17 and 20 exhibiting the greatest cellular efficacy and inhibiting the ALDEFLUOR signal by more than 90% at 1 µM. The generally increased cellular efficacy of this series, and especially of 17 and 20, is consistent with prior observations that many of the most efficacious compounds toward the PEO1 cell line have their in vitro potency skewed toward the 1A2 and1A3 isoenzymes due to the high expression of the 1A3 isoenzyme in these cells [43]. Although all compounds were pre-incubated with the enzyme for two minutes prior to initiation of the enzymatic reaction, variation in the preincubation time for compounds 1, 17 or 20 did not demonstrate any time- dependency for the extent of inhibition, and, thus, these compounds do not appear to react covalently. None of the isoenzymes tolerated a methyl substituent at the R⁷ position (21), likely due to the hydrophilic nature of the ALDH active site in this region, which is exemplified by the interaction of the nitrile group in NCT-506 with the main chain nitrogen atoms of C303 and I304 (Fig. 2C and 2D).

Table 4.

Optimization of Benzimidazole Ring Structure.

			ALDH EC50 (nM) or % Control at 20 µM			CLogP	PEO1 Cell ALDEFLUOR (% inhibition)
No.	R7	R8	1A1	1A2	1A3
6	−H	−H	190 ± 15	240 ± 57	79 ± 30	4.27	50 ± 4% (10 µM)
15	−H	−Cl	230 ± 21	49 ± 2	110 ± 10	4.94	49 ± 6% (1 µM)
16	−H	−F	33%	45 ± 7	100 ± 7	4.40	56 ± 6% (1 µM)
17	−H	−CN	140%	43 ± 2	110 ± 9	4.25	98 ± 2% (1 µM)
18	−H	−OCH3	240 ± 23	55 ± 3	120 ± 4	3.89	75 ± 6% (1 µM)
19	−H	−CH3	98 ± 6	91 ± 8	150 ± 8	4.59	26 ± 5% (1 µM)
20	−H	−C≡CH	160 ± 54	25 ± 11	57 ± 2	4.33	93 ± 4% (1 µM)
21	−CH3	−H	120%	74%	120%	4.70	ND

3.6. Optimization of the benzimidazole N-alkyl substituent

Lastly, we examined how variations in the N- alkyl group impacted this series (Table 5). Consistent with its location near the alcove created by G458 (Fig. 2A and 2B), simple extension of the N-propyl group to N-butyl, eliminated potency toward 1A2 and 1A3 (compound 8 versus 22). However, cyclization of the butyl group to methyl-cyclopropyl restored but did not optimize potency for 1A2 and 1A3 (compound 24 versus 6). This is consistent with the substrate-binding site topography near the alkyl group in the 1A1 complex with compound 4 (Fig. 2A and 2B) where additional length will better fill this region of the binding site in 1A1. However, the asparagine side chain that replaces G458 at the equivalent position in 1A2 and 1A3 (N475 and N469, respectively) fills this structural alcove and precludes additional length at this position of the benzimidazole scaffold. Interestingly, N-alkylation with an unbranched 4-atom ether substituent was detrimental to potency for all isoenzymes (compound 23), whereas an oxetanyl substituent is best optimized for pan-1A potency and retains significant cellular efficacy (compound 25 vs compound 15). It is possible that the CLogP of compound 23, which lacks a Cl at R⁸, is too low for effective binding to the hydrophobic substrate binding site of ALDH1A isoenzymes, since no potent or efficacious compounds possessed CLogP values below 3.8. Thus, desolvation penalties associated with more hydrophilic compounds may dominate the energetics of binding.

Table 5.

Optimization of the benzimidazole N-alkyl substituent.

				ALDH EC50 (nM) or % Control at 20 µM			CLogP	PEO1 Cell ALDEFLUOR (% inhibition)
No.	R2,8	R4,5	R6	1A1	1A2	1A3
8	−H,−H			780 ± 78	2200 ± 200	1800 ± 500	4.72	ND
22	−H,−H			620 ± 59	98%	97%	5.18	ND
6	−Cl,−H	−H,−H		190 ± 15	240 ± 57	79 ± 30	4.27	50 ± 4% (10 µM)
23	−Cl,−H	−H,−H		3300 ± 710	4500 ± 990	1500 ± 1200	3.35	ND
24	−Cl,−H	−H,−H		97 ± 29	390 ± 84	250 ± 10	4.17	ND
25	−Cl,−Cl	−H,−H		83 ± 12	45 ± 6	43 ± 2	3.89	97 ± 2% (10 µM) 29 ± 5% (1 µM)

3.7. Metabolic stability studies

In order to prioritize candidates for initial in vivo studies, selected compounds were evaluated for stability to incubation with mouse liver microsomes (Table 6). The early analog, 4, had a half-life of only 4.3 minutes, indicating rapid oxidative metabolism. Surprisingly, replacement of the potentially labile methoxy group on the phenol by chlorine (6) did not improve stability, suggesting that the phenol ring is not a major site of metabolism. Greater stability was realized by 8-chloro substitution on the benzimidazole ring (15), which increased the half-life of 6 by over 5-fold. A nitrile at the 8-position (17) was also stabilizing, but not as much as chlorine. In addition to the relatively electron-rich benzimidazole ring, the N-alkyl sidechain presents another potential target for oxidative metabolism. Therefore, N- cyclopropylmethyl analog 24 was examined, but showed no improvement in stability over simple N- propyl analog 6, suggesting that the major site of metabolism is also not the alkyl sidechain. Ultimately, optimum metabolic stability was realized by retaining the 8-chloro group of 15 and replacing the N- propyl with the more hydrophilic oxetanylmethyl group (25).

Table 6.

Stability in mouse liver microsomes.

Compound No.	Half-life (min)
4	4.3
6	3.7
15	19
17	11
24	4.3
25	>60

4. Summary

In this study, through selective modification of the original ALDH1A inhibitory compound CM10 [23, 44], we have demonstrated improvements in the potency of the initial hit compound from low micromolar to less than 100 nM for all isoenzymes with our pan-1A inhibitor, best exemplified by compound 25. This same compound has improved target engagement over the starting compound in PEO1 cells, as well as excellent microsomal stability. Compound 17 is another potentially useful compound, showing strong potency for the 1A2 and 1A3 isoenzymes, and is efficacious in cells that lack expression of ALDH1A1, such as the PEO1 cell line. Conversely, compound 22 is selective toward the 1A1 isoenzyme and would prove useful, when paired with compound 17, to evaluate any distinct contributions of the 1A isoenzymes to underlying cell biology in different cell types. While this work was finishing, another group reported a series of benzimidazole-based compounds, but as a fusion with elements of the CM26 series [49, 62], that yielded ALDH1A inhibition characteristics distinct from the series reported here. An interesting feature of our scaffold is the intramolecular hydrogen bond between the benzimidazole nitrogen and the 2-hydroxyl group of the benzyl substituent. The flexibility of this scaffold to achieve distinct selectivity patterns through modest modifications is a key element and has permitted further exploration of how selectivity for ALDH1A isoenzyme can be achieved. The structural analysis of target engagement with ALDH1A1 further supports the key role that G458 plays in determining selectivity for 1A1 [49], as the longer non- cyclized n-alkyl chains (compound 22 versus compounds 24 or 25) confer selectivity for 1A1 over 1A2 or 1A3. Similar to what was observed in our pyrazolo-pyrimidine series [24, 43], the addition of a nitrile group to the aromatic ring in close proximity to the active site cysteine confers considerable selectivity and potency for ALDH1A3 and/or ALDH1A2 (compound 17), whereas more hydrophobic substituents are preferred by ALDH1A1 at this same position (compounds 18, 19 and 20). The molecular explanation for this phenomenon is less clear, but the active sites of 1A2 and 1A3 isoenzymes are slightly more hydrophilic on approach from solvent (Fig. 2), driven by the substitution N for G458 and T for I304 (N475 and T321 in 1A2 or N469 and T315 in 1A3), which may aid initial binding events. It is interesting that a nitrile group, albeit an alkyl nitrile, is a key feature for how NCT-506 engages with active site of ALDH1A1. Thus, nitrile groups appear to be useful targeting agents for ALDH1A isoenzymes, but their precise spatial location within the active site and the local electronic structure within the inhibitor can influence both the potency and selectivity of the compound. Future work will explore further optimization of the N-alkyl substitutions, as well as pursue assessment of their pharmacokinetic properties in mouse models of disease.

Highlights.

• Compound CM10 was identified from a prior high-throughput screen for inhibitors of ALDH1A1.

• Optimized ALDH1A inhibitors show 15-fold increase in potency over prior lead compound.

• Crystallographic studies support key role of Gly458 in ALDH1A1 selective inhibition.