Structural Evidence for Isoform-Selective Allosteric Inhibition of Lactate Dehydrogenase A.

Lactate dehydrogenase A (LDHA) is frequently overexpressed in tumors, thereby sustaining high glycolysis rates, tumor growth, and chemoresistance. High-throughput screening resulted in the identification of phthalimide and dibenzofuran derivatives as novel lactate dehydrogenase inhibitors, selectively inhibiting the activity of the LDHA isoenzyme. Cocrystallization experiments confirmed target engagement in addition to demonstrating binding to a novel allosteric binding site present in all four LDHA subunits of the LDH5 homotetramer.

Introduction

Lactate dehydrogenase (EC 1.1.1.27) consists of four subunits, all of which catalyze the interconversion of lactate and pyruvate. There are two human LDH isoenzymes (LDHA/B) from which five homo- and heterotetramer combinations or isoforms can be inferred (LDH1 to LDH5). LDH5 is a homotetramer consisting of four identical LDHA subunits and has, for a considerable time, been discussed as a tumor marker due to the detection of elevated levels in serum of cancer patients.[1] Besides tissue breakdown and enzyme release from tumors, overexpression and increased activity of lactate dehydrogenases in intact cancer cells directly contribute to tumor burden by fueling the rapid growth of malignant cells,[2] even in the presence of oxygen (glycolytic phenotype). This may also further the acidification of the tumor microenvironment,[3,4] thereby contributing to chemoresistance. Growing evidence also suggests that the rise in LDH levels may compromise the antitumor immune responses of checkpoint inhibitors in certain tumors.[5,6] Overexpression of LDH5, i.e., LDHA isoenzymes, has been found in a wide variety of tumors,[7−10] whereas the expression of the B isoenzyme is found in specific malignancies only.[11,12] Indeed, LDH5 expression is increased by c-myc and HIF1-alpha, both of which are commonly overexpressed in many tumors. Consequently, LDH5 knockdown diminishes tumor cell proliferation under hypoxic conditions and lowers the tumorigenicity of MCF-7 and MDA-MB-231 breast cancer cell lines in vitro as well as HT29 colon carcinoma cells in vivo.[13] Moreover, lentiviral shRNA-mediated knockdown of LDH5 in human hepatocellular carcinoma cells results in increased pyruvate levels that are associated with a rise in cellular apoptosis (due to an intensified mitochondrial ROS production) and a reduced metastatic potential of these tumor cells.[14] Thus, several academic institutions and pharmaceutical companies have tried to identify small-molecule (SMOL) inhibitors of lactate dehydrogenases. Today, quite a number of such molecules are known (see elsewhere[15,8,9] for a review, and others[16−31] for details). Compounds 8(22) and 9(29) comprise the most prominent representatives reported although their development has been stopped preclinically (see Table for more details). However, given the close amino acid homology of 75% sequence identity of LDHA and B, the tissue distribution, and equivalent pivotal roles of the other LDH isoforms (i.e., LDH1 to LDH4), achieving preferential inhibition of LDH5 has been highly challenging. Here, we report the outcome of a biochemical high-throughput screen resulting in the identification of novel and highly selective SMOL inhibitors of LDH5 in vitro. Moreover, the results of cocrystallization experiments aiming at early demonstration of target engagement proved an allosteric binding mode for these LDHA inhibitors.

Table 1

Comparison of Structural Characteristics and LDHA Selectivity of Compounds 3 and 7 with Reference Compoundsa

Potency values were determined in enzymatic assays based on NADH cofactor consumption [NAD(P)H-Glo]. IC50 values represent the means of at least three independent experiments. Known literature values for reference compounds 8 and 9 are included for comparison.

Results

Identification of Selective LDHA Inhibitors

A biochemical high-throughput screening campaign (see the Supporting Information for full details) led to the identification of phthalimide and dibenzofuran derivatives as two novel classes of selective LDHA inhibitors. The phthalimide derivative, compound 3, 4-[(4-{[(5-chloro-2-thienyl)carbonyl]amino}-1,3-dioxo-1,3-dihydro-2H-isoindol-2-yl)methyl]benzoic acid, was identified as a selective LDHA inhibitor with an inhibitory activity of 308 nM. In addition, the dibenzofuran derivative, compound 7, 1-hydroxy-N-(3-{[(7-nitrodibenzo[b,d]furan-2-yl)sulfonyl]amino}phenyl)cyclopropanecarboxamide, represents the second LDHA selective compound class and exhibits an inhibitory activity of 757 nM. For comparison, the IC50 data of the two reference compounds 8 and 9 were determined to be 6.1 and 21.9 nM, respectively (Table ).

Chemistry

Compound 3 was obtained in two steps starting from the commercially available 4-amino-2-benzofuran-1,3-dione 1. Amide coupling with 5-chlorothiophene-2-carbonyl chloride in pyridine and toluene delivering 2 followed by reaction with 4-(aminomethyl)benzoic acid under acidic conditions at 130 °C yielded the desired compound 3 (Scheme ).

Scheme 1

Synthesis of Compound 3

Reagents and conditions: (a) 5-chlorothiophene-2-carbonyl chloride in pyridine, toluene, rt; (b) 4-(aminomethyl)benzoic acid, HOAc, 130 °C.

The second structural class identified was the 7-nitrodibenzofuran-yl-sulfonylamide cluster. Initial optimization resulted in the identification of compound 7. Synthesis of 7 started by amide coupling of commercially available tert-butyl(3-aminophenyl)carbamate 4 with 1-hydroxycyclopropanecarboxylic acid. Subsequent deprotection of 5 with 4 N HCl gave rise to aniline derivative 6, which reacted with commercially available 7-nitrodibenzo[b,d]furan-2-sulfonyl chloride yielding the desired compound 7 (Scheme ).

Scheme 2

Synthesis of Compound 7

Reagents and conditions: (a) 1-hydroxycyclopropanecarboxylic acid, N,N-dimethylformamide, propylphosphonic anhydride solution (T3P, 50% in EtOAc), diisopropylethylamine, rt; (b) HCl in dioxane (4 M), dioxane, rt; (c) 7-nitrodibenzo[b,d]furan-2-sulfonyl chloride in N,N-dimethylformamide, trimethylamine, 4-N,N-dimethylaminopyridine, rt.

Isoform Selectivity and Activity in Cellular Assays

Compounds 3 and 7 completely inhibited lactate production by recombinant human LDHA in vitro in a dose-dependent manner. When 3 and 7 were tested for their ability to also inhibit the activity of recombinant human LDHB, an unanticipated high selectivity toward the LDHA isoform was found. In stark contrast to the reference compounds 8 and 9, neither compound interfered with LDHB activity at all (IC50 > 25 μM; Table ; see the Supporting Information for assay details). Quite a few additional analogs were synthesized in both series in the course of the hit-to-lead process. As can be expected, some of them exhibited comparable activities, whereas others were less active or inactive. Unfortunately, no SAR data can be given for proprietary reasons. We also tested the cellular activities of 3 and 7 by measuring lactate production by determining the changes in lactate levels in the supernatant of the hepatocellular carcinoma cells Snu-398 and the pancreatic carcinoma cells MIA PaCa-2 using a lactate oxidase based enzymatic assay (see the Supporting Information). Reference compounds 8 and 9 inhibited cellular lactate production with IC50s in the single-digit micromolar (2.2 μM on Snu-398; 2.4 μM on MIA PaCA-2) and in three-digit nanomolar (0.8 μM on Snu-398; 0.3 μM on MIA PaCa-2) range, respectively. Being early development compounds, however, neither 3 nor 7 showed inhibition of lactate production in cells, most likely due to their approximately 100- to 300-fold lower biochemical potency (see Table ; literature values). Although these potency differences are somewhat less striking when taking our data for the reference compounds into account (i.e., 10- to 124-fold lowered potencies), clearly, additional SAR efforts will be necessary to achieve cellular activity.

The Novel LDHA Inhibitors Bind to an Allosteric Pocket

Crystallization efforts were initiated to confirm target engagement and to determine the precise binding sites as well as the binding modes of 3 and 7 in order to support further compound optimization. Complexation of LDHA with 3 and 7 at low protein and ligand concentrations and the implementation of a crystal seeding procedure yielded diffracting crystals (see the Experimental Section for details).[32] Surprisingly, when compared to the seed stock, processing of the X-ray diffraction data revealed that crystals containing compound 3 or 7 exhibited a different space group and comprised a unit cell of altered dimensions (Supporting Information: Table S1). At the resolutions given, strong electron density matching the two compounds was identified at a shared site distinct from the orthosteric substrate- and NADH-binding pockets (Figure A). The data not only allow for the determination of this novel binding site but also of their respective binding modes with high confidence.

Figure 1

Novel LDHA inhibitors bind to an allosteric pocket distinct from the NADH- and pyruvate-binding sites. (A) Overview of the LDHA homotetramer in complex with compound 7 (black, sticks). For clarity, two monomers are displayed as surface (upper graphic) and cartoon (lower graphic) (A/green and B/cyan), while two monomers are shown as surfaces (C/light and D/dark gray). The activation loop is positioned in an open conformation (black arc). (B–D) Comparison of monomer from (A) (green) to orthosteric LDH inhibitors and to the ligand-free form of LDHA (gray). (B) Compound 8 (black/sticks), also known as GNE-140, binds cooperatively with NADH to the active site (gray; 4ZVV.pdb). (C) An analog of 9 (black/sticks), an NADH-competitive compound discovered at GSK, binds at the far end of the NADH-binding site where the adenine moiety would interact (gray; 4QSM.pdb). (D) No major structural rearrangements are observed when compared to apo LDHA (gray; 4L4R.pdb).

In fact, the novel binding site is located at the interface between two LDHA monomers, meaning the compounds interact, although to a different extent, with residues from two neighboring subunits. In accordance with full enzyme inhibition, all four binding sites were found to be occupied. Individually, 3 and 7 exhibit virtually identical binding poses in each of the four sites. Examination of all publicly available crystal structures of eukaryotic LDH proteins (sequence identity >40% to hLDHA) revealed that no SMOL ligands have been shown to bind to this site (Supporting Information: Figure S1). Given that both compounds prompt potent enzymatic inhibition, we concluded that this site can be considered as an allosteric LDHA binding site. Overall, the secondary structure elements of the LDHA tetramer were intact, and no dramatic conformational changes were observed within either the cofactor- (NADH) or the adjacent substrate-binding (pyruvate) domains (Figure B–D). The active site loop of the LDHA molecule is positioned in an “open” conformation similar to the LDHA structure in the absence of any substrate and NADH (Figure D).[32] Nevertheless, to accommodate the allosteric inhibitors, minor but similar conformational changes are observed in the two X-ray structures (Figure A).

Figure 2

Structural rearrangements in LDHA upon inhibitor binding. (A) Comparison of LDHA/7 (green and cyan/cartoon; orange/sticks) and a crystal structure of LDHA in complex with NADH/oxamate (gray/cartoon and black/sticks; 1I10.pdb). Minor but key changes can be observed in Helix 7 and Helix 8. (B) Rotation of Helix 7 displaces Tyr172, which is crucial in opening the ligand binding site. (C) Movement of Helix 8 and the retraction of Arg169 may lead to the loss of substrate coordination. For clarity, only certain residues are shown. Co-complex structures were superimposed to each other by simultaneous alignment of both chains.

Structural LDHA Rearrangements upon Inhibitor Binding

Compared to apo LDHA (4L4R.pdb), the most obvious rearrangements are found in and around Helix 7, which upon binding of compounds 3 and 7 rotates out by 6.6 and 7.7°, respectively (Figure B). The relocation of Tyr172 in Helix 7 by approximately 3.5 Å (as measured on the side-chain hydroxyl group) appears to be critical to ligand binding. In addition, Helix 8 straightens out slightly, pushing its central residues ∼1 Å closer to the bound inhibitors (Figure C). Interestingly, changes can also be detected in the positioning of Arg169, which coordinates the LDHA inhibitor oxamate (a pyruvate analog). The apparent retraction of its guanidinium head group, by ∼1 Å, is highly suggestive of the loss of proper substrate coordination as a potential rationale for the observed inhibition of enzymatic activity (Figure C).

Taken together, the described structural changes expose an extended and relatively narrow binding site between two adjacent LDHA monomers. Compounds 3 and 7 bind to this allosteric site with their respective molecular core regions, phthalimide and dibenzofuran, respectively, in an overlapping manner. Interestingly however, their noncore moieties protrude in different directions (Figure A).

Figure 3

Protein–ligand interactions of the two novel allosteric inhibitors. (A) The allosteric inhibitors bind in the same primary pocket but extend in two different directions. (B) Compound 3 exhibits stacking with Tyr172 and extends out of the central pocket toward Lys59. (C) Similarly, compound 7 also exhibits stacking with Tyr172 but rather protrudes away from Chain B. The inhibitor hydrogen bonds to the backbone of Ile177 and Pro75 and to the side-chain hydroxyl of Tyr172.

LDHA–Small-Molecule Inhibitor Interactions

The benzoic acid of 3 extends toward one LDHA monomer (e.g., Chain B), while the hydroxy cyclopropyl phenyl amide of 7 protrudes in the completely opposite direction toward the adjacent monomer (e.g., Chain A). The principal and slightly deeper binding pocket is lined by a combination of hydrophobic (Leu70*, Pro75*, Ile77*, Tyr172, and Leu173) and hydrophilic residues (Gln66*, His67*, Ser69*, Thr74*, Lys76*, Arg169, Gln233, and Ser237). The asterisk denotes residues coming from the neighboring chain, e.g., B. Overall, the bottom of this pocket is slightly more hydrophobic than its rim. Both inhibitors exhibit stacking interactions to Tyr172 on one side and to Gln66* and His67* on the other side (Figure B). The phthalimide moiety of 3 also interacts by a hydrogen bond to the backbone amide hydrogen of Ile77* and orients its benzoic acid toward Lys59*, Met62*, and Val78* (Figure B). In contrast, the sulfonamide of the dibenzofuran 7 interacts with the backbone atoms of Ile177* as well as Pro75* and induces a sharp turn of the molecule in the opposite direction (Figure C). This allows for supplementary interactions to another set of residues (Gly175, Glu176, Gly179, Val180, His181, and Pro182). In addition, Arg268 of a third monomer (e.g., D) extends close to compound 7. However, Arg268 exhibits a different conformation in the complex structure of compound 3, indicating flexibility in the conformation and location of this residue.

Discussion

In this paper, we report selective allosteric inhibitors of the human LDHA isoenzyme with sub-micromolar activity in vitro. Cocrystallization experiments were used to confirm target engagement and to identify the actual binding sites of these phthalimide (3) and dibenzofuran (7) derivatives. The compounds were shown to bind in an overlapping fashion to a novel binding pocket distant from the polar and extended orthosteric substrate/cofactor binding site. All LDHA inhibitors identified so far bind to these “classical” pockets in some way or another.[15] For example, compound 8 blocks pyruvate substrate binding through cooperative binding with NADH, while compound 9 hinders cofactor binding in an NADH-competitive mode (Figure B,C).

However, given the high similarity of LDH active sites, neither 8 nor 9 is particularly isoform-selective and inhibits LDHB as well (Table ). Hence, there is still a need for additional chemical matter that potently and selectively inhibits LDHA to fully take advantage of isoform-specific tissue distribution and disease-related expression patterns. Beyond therapeutic advantages resulting from increased selectivity with regard to closely related proteins, the identification of allosteric inhibitors may offer additional benefits such as avoiding substrate competition effects and influencing LDHA subunit cooperativity.[33] Indeed, allosteric activation by fructose-1,6-bisphosphate (FBP) has been described for the lactate dehydrogenase of Bifidobacterium longum,[34] and the option of allosteric LDH inhibition has been postulated based on studies recording dynamic motions within LDH proteins.[35,36] Additional studies demonstrated the disruption of rate-promoting vibrations (RPVs) upon binding of 2-chloro-N-(3,5-dihydroxyphenyl)acetamide (CPA; selected by docking studies) to a predicted pocket in direct proximity of the active site in human heart LDH (hLDHB).[37] Building on that, Andrews and Dyer have recently identified partial allosteric inhibitors of porcine heart LDH and reported a noncompetitive inhibitory activity for 3-acetamidophenol (3-AP) in the micromolar range in vitro (IC50 = 78 ± 21 μM).[38]

We herein present two isoform-selective LDHA inhibitors with sub-micromolar activities in vitro and demonstrate both target engagement and inhibitor binding to a novel allosteric site with the help of protein–ligand crystal structures. Given that no crystal structures of human LDHA or homologous proteins deposited to date exhibit this binding-competent allosteric pocket in an open conformation (Supporting Information, Figure ), it is tempting to speculate that binding of 3 and 7 may well result from ligand-dependent conformational selection. The selection of such a less favored conformational state seems to be accompanied by either additional subtle conformational[39] or dynamic changes in the substrate and/or cofactor binding pocket[40,41] that result in enzyme inhibition. Whether the LDHA inhibition finally observed actually results from conformational changes or dynamic processes relevant to enzymatic turnover will have to be the subject of further studies.

Interestingly, most residues within ≤5 Å of any of the ligands are identical in LDHB (Supporting Information, Figure ). Hence, the immediate ligand interacting residues are unlikely to be solely responsible for the exceptional selectivity observed. This proposes the idea that access to a similar allosteric site in LDHB may not be allowed due to a restriction or at least a difference in the conformational space adopted by LDHB.

Conclusions

Even though compounds 3 and 7 represent good starting points, further efforts to optimize their potency are clearly required in order to achieve activity in cellular systems in vitro as well as in appropriate in vivo models. Classical medicinal chemistry approaches may particularly benefit from specifically focusing on target residence times as highlighted in the recent literature on LDH inhibitor design.[42] In addition, in silico efforts, in particular molecular docking and molecular dynamics simulations,[43] may prove to be especially helpful with regard to the identification of novel chemical matter and detailed studies of the energy landscapes of LDHA versus LDHB. Revisiting LDHA, quite a classical workhorse in the early days of protein crystallography,[44] may also contribute to a better understanding of allosteric structural changes in multimeric proteins and serve as a test case for the development of improved computational algorithms predicting allostery.[45] Thus, the current findings may contribute to the identification of additional improved allosteric compounds and serve as a starting point for development of selective and clinically efficacious LDHA inhibitors.

Experimental Section

4-[(4-{[(5-Chloro-2-thienyl)carbonyl]amino}-1,3-dioxo-1,3-dihydro-2H-isoindol-2-yl)methyl]benzoic Acid (3)

Step 1. To a solution of 4-amino-2-benzofuran-1,3-dione 1 (0.2 g, 1.23 mmol) in pyridine (1.44 mL) was added a solution of 5-chlorothiophene-2-carbonyl chloride (0.33 g, 1.84 mmol) in a mixture of pyridine and toluene (0.8 mL each). The reaction mixture was stirred at room temperature under an argon atmosphere for 24 h. After evaporation of the solvents, the crude residue (0.760 g, >100%) was used in the next step without further purification. Step 2. 5-Chloro-N-(1,3-dioxo-1,3-dihydro-2-benzofuran-4-yl)thiophene-2-carboxamide 2 (crude product, 0.760 g, 2.47 mmol) and 4-(aminomethyl)benzoic acid (0.373 g, 2.47 mmol) were suspended in acetic acid (25 mL). The reaction mixture was stirred at 130 °C for 20 h. The mixture was allowed to cool to room temperature, and the crystals were sucked off yielding the desired compound 4-[(4-{[(5-chloro-2-thienyl)carbonyl]amino}-1,3-dioxo-1,3-dihydro-2H-isoindol-2-yl)methyl]benzoic acid (3) (0.495 g, 45.4%). 1H NMR (400 MHz, DMSO-d6) δ [ppm] 4.86 (s, 2H); 7.34 (d, J = 4.31 Hz, 1H), 7.45 (d, J = 8.36 Hz, 2H), 7.69 (d, J = 7.35 Hz, 1H), 7.79 (d, J = 4.31 Hz, 1H) 7.85–7.92 (m, 3H), 8.34 (d, J = 8.36 Hz, 1H), 10.25–10.49 (br, 1H), 12.89–13.00 (br, 1H).

1-Hydroxy-N-(3-{[(7-nitrodibenzo[b,d]furan-2-yl)sulfonyl]amino}phenyl)cyclopropanecarboxamide (7)

Step 1. To a solution of tert-butyl(3-aminophenyl)carbamate 4 (1 g, 4.80 mmol) in N,N-dimethylformamide (11.08 mL) were added 1-hydroxycyclopropanecarboxylic acid (0.588 g, 5.76 mmol), propylphosphonic anhydride solution (T3P, 50% solution in ethyl acetate), (5.5 g, 5.046 mmol), and N,N-diisopropylethylamine (2.17 g, 2.92 mL, 168 mmol). The reaction mixture was stirred at room temperature for 66 h. After filtration, the filtrate was purified by HPLC (water, HCOOH/acetonitrile (0.1%)) yielding the desired product tert-butyl (3-{[(1-hydroxycyclopropyl)carbonyl]amino}phenyl)carbamate 5 (0.17 g, 9.9%, 81% pure). Step 2. To a solution of tert-butyl (3-{[(1-hydroxycyclopropyl)carbonyl]amino}phenyl)carbamate 5 (0.17 g, 0.471 mmol, 81% pure) in dioxane (1 mL) was added a solution of HCl in dioxane (4 M, 0.94 mL). The reaction mixture was stirred at room temperature overnight. Due to an incomplete reaction, additional HCl in dioxane (4 M) was added (0.5 mL). Stirring was continued at room temperature for 2 h. The precipitate was filtered, washed with dioxane, and dried at room temperature yielding N-(3-aminophenyl)-1-hydroxycyclopropanecarboxamide hydrochloride 6 (0.125 g, >100%). Step 3. To a solution of 7-nitrodibenzo[b,d]furan-2-sulfonyl chloride (0.06 g, 0.192 mmol) in N,N-dimethylformamide (1 mL) were added triethylamine (0.058 mg, 0.08 mL, 0.577 mmol), dimethylaminopyridine (2 mg, 0.019 mmol), and N-(3-aminophenyl)-1-hydroxycyclopropanecarboxamide hydrochloride (0.073 g, 0.289 mmol, 90%). The reaction mixture was stirred at room temperature overnight. The reaction mixture was treated with N,N-dimethylformamide and water (1 mL each), filtered, and finally purified by HPLC (water, HCOOH/acetonitrile (0.1%)) yielding the desired compound 1-hydroxy-N-(3-{[(7-nitrodibenzo[b,d]furan-2-yl)sulfonyl]amino}phenyl)cyclopropanecarboxamide (7) (45 mg, 47.5%). 1H NMR (400 MHz, DMSO-d6) δ [ppm] 0.86–0.96 (m, 2H), 1.07–1.17 (m, 2H), 6.43–6.56 (br, 1H), 6.79 (br d, J = 7.35 Hz, 1H), 7.10 (t, J = 8.11 Hz, 1H), 7.26 (d, J = 8.11 Hz, 1H), 7.80 (br, 1H), 7.97–8.10 (m, 2H), 8.35 (dd, J = 8.52, 1.52 Hz, 1H), 8.56 (d, J = 8.60 Hz, 1H), 8.71 (d, J = 1.57 Hz, 1H), 8.91 (br, 1H), 9.78 (br, 1H), 10.40 (br, 1H).