Structure-Based Design of Transport-Specific Multitargeted One-Carbon Metabolism Inhibitors in Cytosol and Mitochondria

Multitargeted agents provide tumor selectivity with reduced drug resistance and dose-limiting toxicities. We previously described the multitargeted 6-substituted pyrrolo[3,2-d]pyrimidine antifolate 1 with activity against early- and late-stage pancreatic tumors with limited tumor selectivity. Structure-based design with our human serine hydroxymethyl transferase (SHMT) 2 and glycinamide ribonucleotide formyltransferase (GARFTase) structures, and published X-ray crystal structures of 5-aminoimidazole-4-carboxamide ribonucleotide formyltransferase/inosine monophosphate cyclohydrolase (ATIC), SHMT1, and folate receptor (FR) α and β afforded 11 analogues. Multitargeted inhibition and selective tumor transport were designed by providing promiscuous conformational flexibility in the molecules. Metabolite rescue identified mitochondrial C1 metabolism along with de novo purine biosynthesis as the targeted pathways. We identified analogues with tumor-selective transport via FRs and increased SHMT2, SHMT1, and GARFTase inhibition (28-, 21-, and 11-fold, respectively) compared to 1. These multitargeted agents represent an exciting new structural motif for targeted cancer therapy with substantial advantages of selectivity and potency over clinically used antifolates.


■ INTRODUCTION
Single-target inhibition in cancer is all too often insufficient for tumor eradication.The heterogeneous nature of most cancers enables development of resistance to individual therapeutic agents.Further, currently used anticancer agents are generally not tumor-specific.Rather, they attack normal cells as well as tumors and, as a result, dose-limiting toxicities often limit their clinical effectiveness. 1 Indeed, a general lack of tumor selectivity and consequent toxicities combined with chemotherapy resistance significantly contribute to the failure of most antitumor therapies.

Journal of Medicinal Chemistry
inosine monophosphate cyclohydrolase (ATIC)) (Figure 1). 16,17ll of the clinically used antifolates (Figure 2) are transported by RFC. 12,16Indeed, drug transport by RFC by normal tissues is a major cause of dose-limiting toxicities for all of the clinically used antifolates.Further loss of RFC transport is a common mechanism of antifolate resistance. 7,161][12][13]18,19 Ideally, these agents would show limited uptake by RFC.
Following internalization, cytosolic folates are transported into mitochondria by SLC25A32. 20Although serine, glycine, and formate exchange between mitochondria and cytosol, 6 mitochondrial folates do not exchange with folates in the cytosol. 20In the mitochondria, serine hydroxymethyl transferase (SHMT) 2 catalyzes the conversion of serine and tetrahydrofolate (THF) to glycine and 5,10-methylene THF (Figure 1). 6,215,10-Methylene THF is subsequently metabolized by MTHFD2 (5,10-methylene THF dehydrogenase 2) to 10formyl THF, which is converted to formate by MTHFD1L (5,10-methylene THF dehydrogenase 1-like). 6,21In the cytosol, MTHFD1 (5,10-methylene THF dehydrogenase 1) catalyzes the synthesis of 10-formyl THF from formate and THF.10-Formyl THF is the C1 donor for de novo purine biosynthesis in reactions catalyzed by the GARFTase and AICARFTase activities of the bifunctional enzyme ATIC (Figure 1).10-Formyl THF is also converted to 5,10-methylene THF by MTHFD1 for thymidylate synthase and other reactions including SHMT1 and 5,10-methylene THF reductase.In cancer cells, serine is the primary source of C1 units for the biosynthesis of purines and thymidylate in the cytosol and accounts for >85% of glycine for the synthesis of proteins, purines, heme, and glutathione. 6,21In addition, mitochondrial C1 metabolism from serine is also an important source of NAD(P)H and ATP. 6,21mportantly, serine catabolism is frequently upregulated in cancer, with SHMT2 and MTHFD2 among the most overexpressed metabolic genes in human cancers compared to normal tissues. 22−26 Further, overexpression of SHMT2 is associated with poor prognosis in breast cancer, 27−29 lung adenocarcinoma, 30 pancreatic cancer, 31 gliomas, 32 and gastrointestinal cancers. 33In invasive breast cancer, adrenal carcinoma, and renal cell carcinomas, SHMT2 is likewise overexpressed. 27Collectively, these studies argue that SHMT2 is a promising therapeutic target for a broad range of cancers including late-stage tumors. 21,34owever, the challenges in targeting C1 metabolism can be formidable.These range from limited tumor selectivity (reflecting the importance of particular C1 pathways to both tumors and normal tissues) to the lack of effective tumor targeting via tumor-selective uptake by FRs 18 or PCFT 8 over the ubiquitously expressed RFC. 7 Further, for SHMT2, loss of activity results in a compensatory reversal of SHMT1 catalysis (serine → glycine), 35 thus limiting the metabolic impact.Indeed, an ideal C1 inhibitor can be envisaged to inhibit multiple cellular targets in both the mitochondria and cytosol, while preserving tumor selectivity via uptake by FRs and/or PCFT over RFC.
We recently discovered potent, multitargeted antifolates 1−4, all 5-substituted pyrrolo [3,2-d]pyrimidine compounds (Figure 3). 36The most active compound 1 was transported into tumor cells by RFC and PCFT, and accumulated in both mitochondria and cytosol as polyglutamates which act as potent inhibitors of mitochondrial (SHMT2) and cytosolic (SHMT1, GARFTase, and AICARFTase 36 ) C1 targets. 5Compound 1 showed moderate in vitro antitumor efficacy toward multiple tumor types, including pancreatic cancer, lung adenocarcinoma, and colon cancer cell lines, 5,36 and showed potent in vivo efficacy against both early-and late-stage MIA PaCa 2 pancreatic cancer xenografts. 36Compound 1 is a first-in-class "classical" antifolate agent that inhibits multiple cellular targets in C1 metabolism in both the mitochondria and cytosol.
In this report, we used compounds 1−4 as lead analogues to explore the structure−activity relationships (SAR) for 11 new compounds of this series as a means of optimizing transporter selectivity and increasing multitarget potency and specificity.To optimize and validate target inhibition, we determined the X-ray structures for human SHMT2 in complex with 1 and human GARFTase complexed with 1, 2, and 3.In the design of our  [3,2-d]pyrimidine antifolates with phenyl (1 and 2) and thienyl (3 and 4) L-glutamates and 3−5 bridge carbons. 36ultitargeted C1 inhibitors, we used molecular modeling based on these structures for SHMT2 and GARFTase, along with published X-ray crystal structures of human ATIC, 37 rabbit SHMT1, 19 human FRα, 38 and human FRβ. 39RESULTS AND DISCUSSION Design of Multitargeted Tumor-Selective Antifolates.Rationale.Compounds 1 (4-atom bridge), 2 (3-atom bridge), and 3 (5-atom bridge) (Figure 3) were the first reported classical antifolates to be actively transported into cells by RFC and PCFT and to inhibit mitochondrial and cytosolic C1 metabolic targets. 36In the design of multitargeted single agents, it is important to provide flexibility since this affords the conformational promiscuity required to bind to multiple cellular targets.On the other hand, rigidity restricts flexibility and precludes multitarget attributes. 40or our proposed pyrrolo [3,2-d]pyrimidine compounds (5− 9, 11−16) (Figure 4), we used a "scaffold hopping" approach for the bridge phenyl in the systematic design and synthesis of multitargeted analogues with variations involving the bridge ring (phenyl, thiophene, and fluorine substitutions).We introduced ample flexibility by incorporating chain lengths of 3-, 4-, and 5carbon atoms into the bridge region, which allows the carboxylate and the bicyclic ring scaffold at the two ends of the molecule to adopt appropriate conformations for binding multiple sites on putative protein targets.
Judicious conformational restriction confined to the phenyl ring and the L-glutamate moieties can be used to enhance the potencies of classical antifolates.For instance, Pendergast et al. 41 described conformational restriction of the L-glutamate via a fluorine−hydrogen bond.We previously adopted this approach to 6-substituted pyrrolo [2,3-d]pyrimidine antifolate analogues. 42Thus, introducing fluorine into bioactive molecules results in additional hydrophobic interactions with receptors and transporters, as well as enzyme targets, and also confers beneficial changes in drug pharmacodynamics. 43,44Based on this, we incorporated fluorine into the bridge aromatic ring of five of the proposed pyrrolo [3,2-d]pyrimidine analogues including 1 (Figure 4), to compare effects on biological activities of the fluorine-substituted compounds with the corresponding des-fluoro analogues.1).
In our previous study of 6-substituted pyrrolo [2,3-d]pyrimidine antifolates, 38 we demonstrated substantial inhibition of KB human tumor cell proliferation associated with cellular uptake by FRα and PCFT, accompanied by a significant reduction in RFC transport.In addition, we showed that differences in biological activity could be attributed to varying the C−X bond distances and C−X−C bond angles (X = heteroatom) in the linker. 38 1), structural alterations that are expected to positively impact transport specificity and target enzyme inhibition.Based on this, we designed a series of 5-substituted pyrrolo [3,2d]pyrimidine compounds with 4 bridge atoms to evaluate and compare the effects of replacing the benzylic CH 2 with O (5 and 6), S (7), and NH (8) (Figure 4).
Design of Multitargeted Transport-Specific Inhibitors.An important goal of this study was to optimize the binding of novel 5-substituted pyrrolo [3,2-d]pyrimidine compounds to both cytosolic SHMT1 and mitochondrial SHMT2, while improving tumor selectivity by circumventing RFC transport and increasing transport by other uptake processes (e.g., FRs).For SHMT2, X-ray crystal structures were previously reported by Scaletti et al. 46 in complex with lometrexol (6QVG) or PMX (6QVl), both exceedingly poor inhibitors of this enzyme.
As our aim was to optimize multitargeted single agents, we also performed molecular modeling with putative cytosolic targets in de novo purine biosynthesis using the published crystal structures of human GARFTase (PDB ID: 5J9F) 48 and human ATIC (PDB ID: 1P4R). 37To optimize the transport of our designed analogues, we used the X-ray crystal structures of human FRα in complex with our novel antifolate N-( 4 38 and human FRβ in complex with PMX (PDB ID: 4KN2). 39The docked scores for the docked compounds with putative enzyme and transport targets are summarized in Table S1 (Supplemental Information).
X-ray Crystal Structure of 1 and Molecular Modeling with Human SHMT2.Pyridoxal phosphate (PLP)-loaded SHMT2 was co-crystallized with inhibitor 1 and serine.The SHMT2 crystal structure in complex with 1 and PLP was solved to a resolution of 2.51 Å, (Figure 6).Data processing and refinement statistics (Table S2), along with detailed interactions and maps for the complex (Figure S1), are included in the Supplemental Information.In both pockets of the SHMT2-compound 1 structure, there was an adduct of PLP bound to glycine (labeled PLG) (Figure 6A, represented in purple).From the crystal structures, the fluoro-phenyl bridge in 1 adopts an orthogonal conformation to the pyrrolo [3,2-d]pyrimidine scaffold.The electron density supported modeling 1 in a single conformation in one pocket (pocket A) but in two distinct conformations in the second pocket (pocket B).One of the conformations in pocket B was similar to the conformation in pocket A, suggesting  The fluoro-phenyl ring of 1 adopts three different conformations in the two binding pockets of SHMT2.In pocket A, the fluorine on the phenyl ring and O18 of the amidecarbonyl reside on the same side.As noted above, one conformation in pocket B is similar to the conformation in pocket A. For the other conformation in pocket B, the amide NH (N19) resides on the same side as the fluorine (Figures 7 and S1, Supplemental Information).This hydrogen bond interaction between the fluorine and the amide N19-H may lock the orientation of the L-glutamate tail in a specific conformation and is predicted to enhance the binding affinity of 1.
We used the conformation in pocket A as the prototype to design additional SHMT2 inhibitors.From molecular modeling, it was determined that the energy-minimized conformation of 1 has the fluorine and O18 of the amide-carbonyl on the same side of the aromatic ring (Figure 8).
Human SHMT2 complexed with 1 (PDB ID: 8FJU) was used for in situ modeling of the designed compounds (Figure 4) and docking studies.The docked scores of the proposed compounds are summarized in Table S1 (Supplemental Information).The mitochondrial matrix has an alkaline pH (pH 7.7−7.9) 49,50ompared to the cytosol (pH 7.0−7.4); 50,51for docking studies, the protein and our proposed ligands were prepared accordingly.From the docking studies, the 5-atom-bridged compounds showed preferential binding (−13.3 kcal/mol for 11, −10.2 kcal/mol for 9, and −10.0 kcal/mol for 3) and the 3-to 4-atombridged compounds (1, 2, 4, 12, 13, 14, 15, 16) showed slightly decreased docked scores (∼−9 kcal/mol).Analogues with heteroatoms in the bridge (5−8) were found to have poorer docked scores (−7.5 to −8.6 kcal/mol).We attribute these to the higher energy difference between the energy-minimized conformation and the docked conformation compared to the carbon-atom bridged analogues.The design of 12 with two fluorines at the ortho-positions of the phenyl ring to the carbonyl of the L-glutamate tail was based on the two distinct conformations of 1 bound to SHMT2 (Figure 6).As fluorine was observed to adopt positions with both the O18 and N19 in a syn orientation, we designed 12 with the ability to adopt both interactions (by providing two fluorines in one molecule).
Molecular Docking with SHMT1.For docking studies in SHMT1, 19 the X-ray crystal structure of rabbit (O.cuniculus) SHMT1 crystallized with 5-formyl THF tri-glutamate (PDB ID: 1LS3) 19 was used (Figure S2, Supplemental Information).From the conformation of the bound ligand, the bridge along with the phenyl adopts an orthogonal conformation to the pteridine scaffold.In our molecular docking studies with the pyrrolo [3,2d]pyrimidine antifolates, the 5-atom bridged compounds bound in a similar conformation to 5-formyl THF tri-glutamate.From induced-fit docking, 1 adopts an orthogonal conformation in the binding pocket.Both the 2-NH 2 and the N3 of the bicyclic ring form hydrogen bonds with the backbone carbonyl of Gly125; the 4-CO of the pyrimidine ring forms a hydrogen bond with the backbone NH of Leu127.The bridge remains within a goodcontact proximity (range ≤ 4 Å) 52 to make hydrophobic interactions with Tyr64, Leu121, Pro258, Phe131, and Leu127.Both the αand γ-carboxylic acids of the L-glutamate of 1 form salt bridges with Lys134; the γ-carboxyl forms a water-mediated hydrogen bond with Ser254 as well.
For 3 and 9, as well as related compounds, the interactions observed for the 5-formyl THF tri-glutamate scaffold were retained in the docked models (Figure S2, Supplemental Information).The N1 makes an interaction with the side chain of Asn347; the 2-NH 2 interacts with the side chain of Asn347 and the backbone of Leu121 and the 3-NH interacts with the backbone of Gly125.Further, similar to 5-formyl THF tri-glutamate, the bridge aromatic ring (phenyl in 11 and thiophene in 3) makes π−π interactions with Tyr64.The L- glutamates of both 9 and 3 are oriented differently from the corresponding tri-glutamate of the 5-formyl THF ligand, due to solvent exposure of the tri-glutamate linker and the presence of a free α-COOH of the first L-glutamate.For 9 and 3, the α-COOH makes ionic interactions with Lys134B and the γ-COOH makes ionic interactions with both Lys134A and Lys134B.The 3-atom (2, 6, and 13) and 4-atom (1 and 14) bridged analogues maintained similar interactions with appropriate conformational adjustments due to a change in the bridge aromatic ring and bridge length (Figure S2, Supplemental Information).All of these molecules retained the orthogonal bound conformation observed for 5-formyl THF tri-glutamate and the other 5-atom bridged compounds.Docking studies of the two 5-atom bridged analogues 9 and 3 showed somewhat better docked scores (−12.3 and −11.9 kcal/mol, respectively) than for the 3-and 4atom bridged compounds (−10.9 to −9.5 kcal/mol) (Table S1, Supplemental Information).
X-ray Crystal Structures and Molecular Modeling with GARFTase.We previously reported a GARFTase crystal structure in complex with a 6-substituted pyrrolo [  As our prior study showed that the 5-substituted pyrrolo[3,2d]pyrimidine 5-atom bridge analogue 3 is a potent inhibitor of GARFTase (K i = 0.33 ± 0.22 μM), 36 we solved the GARFTase crystal structure in complex with substrate β-GAR and 3. We also solved the GARFTase crystal structures in complex with the related compounds 1 (4-atom bridge with intermediate inhibition) and 2 (3-atom bridge with no detectable inhibition up to 150 μM) for in situ modeling of the designed compounds in this study.GARFTase crystal structures in complex with 3, 1, and 2 were solved to resolutions of 2.17, 2.08, and 2.98 Å, respectively (Figure 9, PDB ID 8FJX, 8FJW, 8FJY, respectively).Data processing and refinement statistics (Table S2), as well as detailed interactions and maps for the complexes (Figures S3− S5), are provided in the Supplemental Information.
From the crystal structure of 3 in GARFTase, the fused pyrrolo [3,2-d]pyrimidine ring interacts with the backbone of Leu899, Glu948, and Asp949 through H-bonds, and the 7-CH interacts with the backbone carbonyl of Arg897 through an aromatic H-bond.The five-atom bridge resides in proximity to Asn913, His915, Lys923, Gly924, His944, Val946, and Val950 to make hydrophobic interactions.The α-COOH of the L- glutamate side chain of 3 makes two hydrogen-bond interactions with the protonated side chain of Arg871 and the backbone NH of Ile898.The amide-carbonyl oxygen forms a water-mediated hydrogen bond with the side chain and NH backbone of Ser925.Moreover, the α-COOH of 3 forms a salt bridge with the side  chains of Arg871 and Arg897, while the γ-COOH remains solvent-exposed.
A comparison of the compound 3 GARFTase crystal structure with the crystal structures for GARFTase with 1 and 2 suggested that analogues with a 4-and 5-atom bridge show a preference in binding to the pocket over analogues with a 3-atom bridge.With 1 (4-atom bridge), the N1 makes a hydrogen bond with the backbone NH of Leu899, the 2-NH 2 with the backbone NH of Ser900, and the N3 with the backbone carbonyl of Asp949 and Ala947.The 4-CO of 1 makes a hydrogen bond with the backbone NH of Asp951 and a water-mediated hydrogen bond with Asp940.The amide NH of the L-glutamate of 1 makes a hydrogen bond with Met896.The α-COOH of 1 forms a salt bridge with Lys844 and the γ-COOH forms multiple hydrogen bonds with water molecules.Although 1 and 3 inhibit GARFTase in vitro (at μM and nM concentrations, respectively), 2 with a 3-atom bridge linker did not inhibit GARFTase up to 150 μM. 36Evaluation of 2 with GARFTase showed that many of the polar contacts between the inhibitor and the enzyme binding pocket were substantially longer, to the point that some of these polar contacts do not likely contribute to binding.Specifically, most contacts shared in the 3 and 2 crystal structure complexes showed 0.2−0.8Å longer contact distances in the 2 complex (e.g., polar contacts with backbone atoms of Met896, Leu899, Ala947, and Glu948, and with sidechain atoms of Arg871 and Arg897; cf.contact distances in Figures S4 and S5).Effectively, the 3-atom bridge in 2 is too short to simultaneously allow for optimal polar contacts with GARFTase at both the pyrrolo [3,2-d]pyrimidine and L- glutamate moieties.
Based on inhibition profiles and evaluation of our crystal structures, GARFTase complexed with 1 (PDB ID: 8FJW) was used for in situ modeling of the designed compounds (Figure 4) and docking studies.Compounds 1 (4-atom bridge), 7 (4-atom bridge with a sulfur heteroatom), 9 (5-atom bridge), 14 (4-atom bridge), and 15 (4-atom bridge) showed docked scores ranging from −16.0 to −14.4 kcal/mol (Table S1, Supplemental Information).In the docked pose of 15, both the N1 and the 2-NH 2 interact with Leu899; the 2-NH 2 makes an additional interaction with Glu948, while the 4-oxo forms a watermediated H-bond with Glu948.In the L-glutamate moiety, the α-COOH forms a network of hydrogen bonds with the backbone NH of Arg897 and the side chain of Arg871.The γ-COOH interacts with the side-chain NH groups of Arg897 and Lys844.The thiophene ring of 3 forms nonspecific hydrophobic interactions in the pocket.These interactions were unique from those in the solved GARFTase-compound 3 crystal structure.The remaining analogues showed similar docked poses and interactions to 3 in the GARFTase crystal structure.
Molecular Modeling with ATIC.ATIC is a bifunctional enzyme (AICARFTase/inosine monophosphate cyclohydrolase) in de novo purine biosynthesis, wherein the AICARFTase domain utilizes 10-formyl THF to convert AICAR to formyl-AICAR (Figure 1). 53The crystal structure of human ATIC in complex with the antifolate-based inhibitor N-[(S)-(4-{[(2amino-4-hydroxyquinazolin-6-yl)dihydroxy-λ-4-sulfanyl]a m i n o } p h e n y l ) h y d r o x y m e t h y l ] -L -g l u t a m i c a c i d (BW1540U88UD) (PDB ID: 1P4R) 37 was used for molecular docking studies.This structure was preferred over the crystal structure of ATIC bound to a nonclassical antifolate N-( 6 54 because of the ligand-structure similarity of the classical antifolate to our designed analogues. The docked scores for the 4-atom bridge pyrrolo[3,2d]pyrimidine compounds 14 (−14.4kcal/mol), 16 (−13.9kcal/mol), 1 (−12.9kcal/mol), and 5 (−12.8kcal/mol) suggested a preference for binding to ATIC over the 5-atom bridge compounds 11 (−11.0kcal/mol), 9 (−11.1 kcal/mol), and 3 (−11.7 kcal/mol) (Table S1, Supplemental Information).This difference reflects the loss of interaction between the amino acids in the binding pocket and the L-glutamic acid moiety of the 5-atom bridge compounds (Figure S6, Supplemental Information).The thiophene ring in the 4-atom bridge in 14 and 16, and the phenyl ring in 1 make π−π interactions with the side chain of Phe315 in ATIC.These π−π interactions with Phe315 were absent in the 5-atom bridge compounds as the bridge aromatic (phenyl/thiophene) ring is shifted away (closer to the solvent) in the bound conformation due to the increased bridge length.In the 4-atom bridge compound 15, a π-cation interaction with the side chain of Lys358 is observed.A difference in the orientation of the L-glutamate between the 4-and 5-atom bridge compounds also impacts the interaction with the ATIC protein.The 3-atom bridge analogues (2, 4, and 13) showed differences in molecular docking including a lack of the π−π interaction with Phe315.In addition, due to a shorter bridge length, the L-glutamic acid does not actively interact with the amino acid side chains in the binding pocket.All of the additional analogues (with minor positional adjustments) maintained similar interactions of the pyrrolo [3,2-d]pyrimidine scaffold with the enzyme (Figure S6, Supplemental Information).
Compound 1 in the docked pose of human FRα binds in the folate-binding cleft (Figure S7, Supplemental Information).The 2-NH 2 and 4-oxo moieties of 1 interact with amino acids in the binding pocket similar to the crystallized ligand 10.The 2-NH 2 and 3-NH groups interact with Ser147, and the 4-oxo forms a hydrogen bond with the side-chain NH of Arg103.The pyrrolo [3,2-d]pyrimidine scaffold is sandwiched between the side chains of Tyr85 and Trp171, similar to that seen with the pyrrolo [2,3-d]pyrimidine ring of 10 in its bound conformation.The L-glutamate moiety of 1 is oriented similar to the corresponding L-glutamate in 10. 55 The α-COOH of 1 forms a network of hydrogen bonds that involves the backbone NH of Gly137 and Trp138, and the side-chain NH of Trp140.The γ-COOH of 1 interacts with the side-chain NH moieties of Lys136 and Trp102 (not labeled), similar to that observed in the corresponding L-glutamate portion of 10 in the crystal structure with FRα. 55The ortho fluorine-substituted phenyl ring of 1 forms hydrophobic interactions with Trp102 and a π-cation interaction with protonated Arg103.The other 4-carbon-atom bridge analogues (12, 14, 15, and 16) (Figure S7, Supplemental Information) showed similar interactions in the binding pocket and the best docked scores (−18.2 to −20.5 kcal/mol) (Table S1, Supplemental Information).The compounds with 4 bridge atoms including heteroatoms (7, 8, and 5) showed decreased To analyze binding to FRβ, the proposed analogues were docked in the X-ray crystal structure of FRβ bound to PMX (PDB 4KN2; Figure S8, Supplemental Information). 39The results predict that compounds with 3-to 4-carbon bridge compounds are preferred for FRβ uptake.Based on their docked scores (Table S1, Supplemental Information), 5, 13, and 15 are predicted to be among the top three compounds for FRβ transport.Compound 5 contains an oxygen heteroatom in the bridge adjacent to the phenyl ring that did not make any contacts in the binding pocket, nor did it interfere with the ability of the ligand to mimic the PMX-bound conformation. 39The 2-NH 2 and 4-oxo moieties of 5 interact in a similar manner to the corresponding groups of the PMX ligand, with the 2-NH 2 and 3-NH interacting with Ser147, and the 4-oxo forming hydrogen bonds with the side-chain NH of Arg119 and His151.The pyrrolo [3,2-d]pyrimidine scaffold is sandwiched between the side chains of Tyr101 and Trp187, similar to that seen with the pyrrolo [2,3-d]pyrimidine scaffold of PMX in its bound conformation.The L-glutamate moiety of 5 is oriented similar to that of PMX with FRβ. 39The α-COOH of 5 forms a network of hydrogen bonds that involve the backbone NH of Gly153 and a water-mediated H-bond with Asp155.The γ-COOH interacts with the side-chain NH groups of Arg152 and Gln116.The phenyl ring of 5 forms hydrophobic interactions with Trp150 and π−π interactions with Trp118 and Trp156.The hydrophobic 3-C part of the bridge of 5 forms nonspecific hydrophobic interactions in the binding pocket.The other analogues (1, 3, 12, and 14) show similar interactions to 5. Compound 3 with a 5-atom bridge makes appropriate adjustments in the linker conformation so as to retain the pyrrolo [3,2-d]pyrimidine and L-glutamic acid moieties in the same positions as in 5 and to preserve all of the molecular interactions with FRβ.The docked scores and figures of the docked poses of the proposed compounds are included in Table S1 and Figure S8 (Supplemental Information).
Summary of Drug Design.Based on our determinations of the crystal structures and extensive molecular docking of our proposed compounds, we identified eleven 5-substituted pyrrolo [3,2-d]pyrimidine analogues with the potential to bind to FRs and critical enzyme targets (SHMT2, SHMT1, GARFTase, AICARFTase) in mitochondria and cytosol, resulting in potent and selective antitumor efficacies.These provide a compelling rationale for their synthesis and biological characterization.Not surprisingly, comparison of the docked poses of 1 with multiple protein targets showed that the inherent flexibility of our compounds is an essential design element required for the desired tumor-specific transport and multienzyme targeting (Figure 10).
Chemistry.The intermediates (18a and b, Scheme 1) for compounds 5 and 6 were synthesized via an S N 2-like attack on ethyl 4-hydroxybenzoate (17) following a modified version of the reported method. 56Acetonitrile was used instead of N,Ndimethylformamide (DMF) as a solvent to avoid evaporation at higher temp or the extensive workup procedure. 57Compounds 18a and 18b were converted to the mesylate derivatives and then to the respective iodides 19a and 19b using the Finkelstein reaction.
The synthesis of 7 commenced with the Fisher esterification of 4-mercaptobenzoic acid (20) to afford 21 (Scheme 2). 58The substituted thiophenol 22 was obtained via an S N 2 attack of 4mercaptobenzoate ( 21) and Cs 2 CO 3 was used instead of K 2 CO 3 as it is a better counterion for sulfur. 59The alcohols were converted into iodides, following a similar method to that described for 19a and 19b (Scheme 1).
The intermediate for 8, methyl 4-((3-hydroxypropyl)amino)benzoate (25, Scheme 3), was synthesized using methyl 4iodobenzoate (24), CuI (as catalyst), L-proline, and 2-amino ethanol in dimethyl sulfoxide. 59,60Five equivalents of 2-amino ethanol were used to function as both base and nucleophile.The usual method to convert alcohols to iodides showed multiple products on thin-layer chromatography (TLC) analysis, and the yields were poor.Hence, a one-pot Appel reaction was used to obtain 27. 56This afforded a much better yield.
While syntheses of compounds 1 and 11 were published earlier, 36 improved synthetic approaches are described in Scheme 4. Syntheses of 1, 11, 12, and 13 (Scheme 4) started with a palladium-catalyzed Sonogashira coupling of 28a and b with the appropriate alkyne alcohols to afford 29a−d.Catalytic hydrogenation of 29a−d afforded the saturated alcohols 30a− d. 48Compounds 30a−d were converted to their respective iodides 31a−d using the Appel reaction as in Scheme 3. The Nalkylation using iodides 31a−d afforded the N-5-substituted pyrroles 32a−d.The crude N-substituted pyrroles 32a−d were directly subjected to condensation with 1,3-bis-(methoxycarbonyl)-2-methylthiopseudourea and subsequent saponification at 55−65 °C to afford the free pteroic acids 33a−d. 61Peptide coupling of 33a−d with L-glutamic acid diethyl esters using HATU as the coupling reagent gave 34a−d and subsequent saponification afforded the target compounds 13, 1, 11, and 12, respectively.Previous coupling of the pteroic acids 33b and 33c with the L-glutamate ester for the synthesis of 1 and 11 utilized CDC as a coupling reagent and NMM as a base, 36 affording 34b and 34c in 43 and 32% yields, respectively.With the modified method using HATU coupling and N,Ndiisopropylethylamine (DIPEA) as a base, yields were improved for 34b and 34c to 65 and 69%, respectively.
Compounds 14 and 5−8 were synthesized using the synthetic procedure described in Scheme 5. Following the palladiumcatalyzed Sonogashira coupling of 28c with 3-butyn-1-ol to 29e and catalytic hydrogenation of 29e resulted in the formation of 30e.Further conversion of the alcohol of 30e to the iodide (31e) used the same two-step method described in Scheme 1. N-Alkylation using iodide (31e) resulted in the crude Nsubstituted pyrrole (32e).32e was saponified using sodium hydroxide at 55 °C to afford the free pteroic acid 33e. 61Peptide coupling of 33e using CDMT as the coupling reagent with L- glutamic acid diethyl ester gave 34e.Subsequent saponification of 34e afforded the target compound 14.

X-ray Crystal Structures of 5-Substituted Pyrrolo[3,2d]Pyrimidine
Antifolates with SHMT2 and GARFTase.An SHMT2 crystal structure in complex with 14 was solved to a resolution of 2.47 Å (Figure 11; PDB ID: 8FJT).PLP-loaded SHMT2 was co-crystallized with 14 and serine.In pocket A of the 14 structure, an adduct of PLP bound to glycine (Figure 11A; PLG, shown in yellow) was modeled; in pocket B, PLP bound to Lys280 (Figure 11B; LLP, represented in salmon) was modeled.Detailed SHMT2: inhibitor contacts are provided in Figure S9 (Supplemental Information).Although 1 exhibits one conformation in pocket A and two conformations in pocket B (Figure 6), the electron density for 14 adopts only one conformation in each binding pocket.From the crystal structures, the bridge with the fluorothiophene scaffold in In one pocket (pocket A) of the SHMT2 binding site in complex with 14, the binding of the L-glutamate tail in 14 is mediated by polar contacts with the side chain of Lys103 (O24 of the γ-COOH) via salt bridge formation, while the α-COOH has no polar contacts with the protein (Figure S9, Supplemental Information).In the other pocket (pocket B), both carboxylates remain solvent-exposed.For compound 14, the hydrogen bond between the fluorine and the amide NH (N19) is observed in the SHMT2 bound conformation (Figure 12B,C).This interaction partially restricts the orientation of the L-glutamate tail in a specific conformation and is predicted to enhance the binding affinity.From molecular modeling, it was found that the energy-minimized conformation of 14 coordinates the fluorine  and O18 of the amide-carbonyl on the same side of the aromatic ring (Figure 12A).
For comparison to the initial GARFTase crystal structures of 1, 3, and 2 (Figures 9; S3−S5, Supplemental Information), we determined the structure of GARFTase in complex with the newly designed analogue 14 (Figures 13 and S10 S4).The binding of the L-glutamate is mediated by conserved contacts with the side chains of Arg871 and Arg897, as well as the backbone atoms of Ile898 and Met896 (seen in all structures).From this set, only 14 shows contact between the side chain of Arg897 and the γ-COOH (O25), rather than the α-COOH as observed with the 1, 2, and 3 structures.
Biological Evaluation.In Vitro Validation of Intracellular Targets with Isolated C1 Metabolic Enzymes.We measured inhibition kinetics of the pyrrolo[3,2-d]pyrimidine antifolates 5−9 and 11−16 compared to compounds 1−4 from our previous report, 36 with purified enzyme preparations of SHMT1, SHMT2, GARFTase, and ATIC to calculate inhibition dissociation constants (K i s) (Table 2).As expected, the compounds inhibited both SHMT2 and SHMT1 with affinities spanning ranges of 1331-and 304-fold, respectively.The potencies for the top (11 and 9, respectively) and bottom ( 8) compounds (Table 2) for each target paralleled the calculated docking scores (Table S1, Supplemental Information).The overall rank order for inhibition of SHMT2 For both SHMT1 and SHMT2 inhibition, compounds with a 5-atom bridge were preferred, although the bridge aromatic moiety was also a factor.Against SHMT2, 11 (5 atom, 2F phenyl bridge) showed a 28-fold increased potency over lead analogue 1 (4-atom, 2F phenyl bridge).This was followed by 3 (5-carbon, thiophene bridge) and 9 (5-carbon bridge, phenyl bridge) with a 7.5-and 5.6-fold increase in potency, respectively, over 1.A 2fluoro-phenyl bridge (11) increased the inhibition potency by 2fold over a phenyl bridge (9), although replacement of the bridge phenyl with a 2,5-thiophene (cf. 3 and 9) had no impact on the inhibitory activity.
A slightly different inhibition profile emerged with SHMT1, for which the most potent inhibitor was 9 (21-fold more potent than 1), followed by 3 (10.4-foldmore potent than 1) and 11 (5.68-fold more potent than 1).Insertion of heteroatoms (O, S, or N) in the 4-atom bridge (i.e., 5, 7, or 8) had varied effects.Oxygen (5) afforded similar inhibition of SHMT1 and SHMT2 to the carbon isostere 15.Sulfur (7) and nitrogen (8) were generally detrimental to inhibition of both SHMT1 and SHMT2 compared to 15. Insertion of oxygen in the 3-C bridge (6) was also detrimental to the inhibition of both SHMT1 and SHMT2.Although replacement of a phenyl (2) in the 3-carbon bridge series with a thiophene (4) decreased the potency for both SHMT2 and SHMT1, this did not extend to the 4-carbon (15 and 16) and 5-carbon (9 and 3) bridge compounds for which the activities were similar.Likewise, the impact of aromatic fluorine substitutions (13, 1, 11, 14) on inhibitory potencies compared to the nonfluorinated counterparts (2, 15, 9, and 16) was nominal.
For ATIC, K i values spanned an ∼18-fold range with the most potent inhibitor 14 (4-carbon, thiophene bridge) showing a K i of 0.93 μM (4-fold better than 1).However, the differences in inhibitory potencies for the remaining compounds were modest with no particularly significant structural motif.The three compounds with the best K i values (14, 15, and 16) for ATIC (Table 2) all showed the best docked scores (Table S1, Supplemental Information).
Taken together, the in vitro enzyme inhibition assays further confirm multitargeting of SHMT1, SHMT2, GARFTase, and ATIC and establish a robust SAR for the pyrrolo [3,2d]pyrimidine compounds.We also tested the compounds as inhibitors of MTHFD2; these were all inactive up to 200 μM.
The design premise of providing conformational flexibility to allow attachment to multiple targets was born out in the attachment of single agents to different target proteins in quite different conformations made possible by the inherent flexibility of the molecule.This flexibility is illustrated in Figure 14 with compound 1 as an example with the bound crystal structures or  When dealing with multitargeted single-agent inhibitors, it is unlikely that a particular analogue will have the best activity in all targets.Since the intracellular targets SHMT2, SHMT1, GARFTase, and ATIC have different binding sites and hence have different structural requirements including conformations for optimal binding as illustrated in Figure 14, a single analogue may not have the best inhibitory potencies toward all four target enzymes.Accordingly, the analogue with the best balance of inhibitory potencies in all four targets would be considered the most promising.In our series, 9 would be the most promising compound.
Antiproliferative Effects and Antitumor Activities of 5-Substituted Pyrrolo [3,2-d]pyrimidine Analogues.We measured the antiproliferative activities of the 11 novel pyrrolo [3,2d]pyrimidine compounds compared to 1, 2, 3, and 4 from our prior study 36 in relation to the principal mechanisms of (anti)folate transport including the relative contributions of RFC, PCFT, FRα, and FRβ.We initially tested the compounds against a panel of isogenic Chinese hamster ovary (CHO) cell lines engineered from transporter (RFC/FR/PCFT)-null    MTXRIIOua R 2−4 (R2) cells, 62 to express human RFC (PC43-10 cells), PCFT (R2/PCFT4 cells), FRα (RT16 cells), or FRβ (D4 cells). 63−65 By this assay, growth inhibition compared to R2 cells is a direct reflection of transport activity. 63,64he experimental design involved incubations of the CHO cell lines with the pyrrolo [3,2-d]pyrimidine compounds over a range of concentrations for 96 h, after which cell numbers were quantified with a fluorescence-based metabolic assay. 63The results are summarized in Table 3.By this assay, none of the analogues impacted the proliferation of transporter-null R2 cells up to 1000 nM.This establishes an absolute requirement for transport for drug activity.For RFC-expressing PC43-10 cells, 1, 2, and 13 showed inhibition (IC 50 ∼ 42−200 nM), confirming RFC transport, whereas the additional 12 compounds were inactive up to 1000 nM, indicating a lack of RFC transport.Toward R2/PCFT4 cells, growth inhibition was in rank order, 13 > 2 > 1−4−6, with the remaining compounds showing no inhibition (Table 3).
We compared the in vitro potencies of the 11 novel pyrrolo [3,2-d]pyrimidine compounds to compounds 1−4 against KB nasopharyngeal carcinoma cells, characterized by the expression of RFC, PCFT, and FRα 63,64,66 (Table 3 and Figures 15 and S11, Supplemental Information), and HPAC pancreatic adenocarcinoma cells which express RFC and PCFT but not FRs (Table 3). 21Toward KB cells, all compounds were potently inhibitory within a narrow range, from 1.3 nM for 1 to 31. 5 nM for 7.In the presence of 200 nM folic acid, the inhibitory effects were again abolished, although at higher concentrations for some of the analogues (e.g., 1) folic acid was less effective (Figures 15 and S11, Supplemental Information).While this likely reflects uptake by other uptake systems including PCFT, the association between uptake by FRα (i.e., inhibition of RT16 cells) or PCFT (inhibition of R2/PCFT4; Table 3) and KB sensitivity or resistance was inexact.Toward HPAC cells, the potency of compound 13 was similar to that for compound 1 (IC 50 values of 166 and 196 nM, respectively), followed by compounds 2 (310 nM) and 3 (746 nM), and compounds 5, 9, 11, and 14 (ca.1200−1600 nM).The remaining compounds showed modest inhibition (4, 6, 13, 16; IC 50 's from ∼2600 to 4600 nM) or no activity (7, 8, 12; >5000 nM) toward HPAC cells.All of the pyrrolo[3,2-d]pyrimidine compounds were completely inert toward human pancreatic normal epithelial (HPNE) cells, 67   Growth inhibition assays were performed for CHO sublines engineered to express human RFC (PC43-10), FRα (RT16), FRβ (D4), or PCFT (R2/PCFT4), for comparison with transporter-null (R2) CHO cells.Inhibition assays were also performed in KB (nasopharyngeal carcinoma; express RFC, FRα, and PCFT) and HPAC (pancreatic adenocarcinoma; express RFC and PCFT) cells and HPNE cells (human pancreatic normal epithelial cells; express RFC).For the RT16, D4, and KB cells, the experiments were performed in folate-free RPMI 1640 with dialyzed fetal bovine serum and antibiotics with 2 nM leucovorin; 63 for the PC43-10 and R2/PCFT4 CHO experiments, 25 nM leucovorin was provided. 63,64For the RT16 and D4 cells, growth inhibition assays were performed in the presence and the absence of 200 nM folic acid (the latter is not shown).Proliferation inhibition assays were performed using the HPAC pancreatic adenocarcinoma cells and HPNE (human normal pancreatic epithelial) cell lines in folate-and glycine-free RPMI supplemented with 10% dialyzed fetal bovine serum and antibiotics, and 25 nM leucovorin. 5The data shown are mean values from 3−10 experiments (±standard errors in parentheses).Results are presented as IC 50 values, corresponding to the concentrations that inhibit growth by 50% relative to cells incubated without drug.3][64][65]68,69 treated KB human tumor cells cultured in serine-and glycinefree media with the pyrrolo[3,2-d]pyrimidine compounds (4−9, 11−16) to test for the rescue by added nucleosides and glycine. Results re compared to those for AGF94, a pyrrolo [2,3d]pyrimidine antifolate that inhibits GARFTase without effects on SHMT2, 69 and to compounds 1, 2, and 3, documented inhibitors of mitochondrial C1 metabolism and de novo purine biosynthesis in human tumor cells.36 The results are shown in Figures 15 and S11 (Supplemental Information).Inhibition by all of the compounds was unaffected by thymidine alone (10 μM), and there was no impact on cell proliferation of added thymidine (not shown) above the incomplete protection seen with added adenosine (60 μM) alone.In contrast, inhibition by AGF94 was completely reversed by adenosine.Although glycine (130 μM) alone had no effect on growth inhibition, glycine combined with adenosine was completely protective for all of the pyrrolo[3,2-d]pyrimdine compounds in Table 3.This pattern of metabolite protection is diagnostic of mitochondrial C1 inhibitors.36 5-Aminoimidazole-4-carboxamide (AICA) was partly protective from growth inhibition without or with glycine, reflecting inhibition of AICARFTase, the 2nd folate-dependent step in purine biosynthesis.13,36,64 These results suggest that pyrrolo[3,2-d]pyrimidine compounds 5−9 and 11−16, like 1, 2, and 3 previously, 36 are inhibitors of mitochondrial C1 metabolism along with de novo purine biosynthesis in the cytosol, results entirely consistent with the in vitro studies with isolated C1 enzymes (Table 2).

■ CONCLUSIONS
We previously described multitargeted 5-substituted pyrrolo-[3,2-d]pyrimidine antifolates 1, 2, 3, and 4. 36 The lead compound of this series 1 was transported into tumor cells by RFC and PCFT (FR uptake was not evaluated). 5Compound 1 accumulated in the cytosol and mitochondria whereupon it inhibited C1 metabolism at critical mitochondrial (SHMT2) and cytosolic (SHMT1, GARFTase, AICARFTase) targets. 5,36nalogue 1 showed in vitro antitumor efficacy toward pancreatic cancer, lung adenocarcinoma, and colon cancer cell lines, with promising in vivo efficacy against early and late-stage MIA PaCa 2 pancreatic cancer xenografts. 5,36ompound 1 is a first-in-class classical antifolate with a unique spectrum of multitargeted inhibitory activity including SHMT in mitochondria and de novo purine biosynthesis in the cytosol. 36SHMT2 is a bona fide tumor target with significant impacts on glycine and formate availability for cell proliferation and redox homeostasis. 21,34,70There are compelling arguments for targeting de novo purine biosynthesis for cancer, as antipurine drugs inhibit tumors independent of wild-type p53 status, 71 show selectivity based on loss of purine salvage 72 and suppress mTOR signaling. 73he major goal of this investigation was to systematically interrogate the structure−activity profiles for pyrrolo [3,2d]pyrimidine compounds related to compound 1 and similar compounds 2−4 from our prior study. 36We explored the impact of modifications of the bridge (length, heteroatom substitutions) and the bridge aromatic ring (phenyl, thiophene) including the effects of fluorine substitutions.To facilitate our SAR analysis, we determined the X-ray crystal structures of human SHMT2 complexed with 1 and 14, and the X-ray crystal structures of human GARFTase in complex with 1, 2, 3, and 14.We also used molecular modeling based on our SHMT2 and GARFTase structures, as well as published X-ray crystal structures of ATIC, 37 SHMT1, 19 FRα, 38 and FRβ, 39 to design our multitargeted C1 inhibitors.We tested 11 new analogues and compared results to those for 1, 2, 3, and 4 using a range of biological assays to measure transport and target engagement, with antitumor validation by in vitro efficacy studies in KB nasopharengeal tumor cells and HPAC pancreatic cancer cells, with parallel assays with normal (HPNE) cells.
In general, in vitro inhibition of SHMT1 paralleled the inhibition of SHMT2, as expected given the close homology between these proteins (66% sequence identity).In terms of SHMT2 inhibition, compounds with a 5-carbon bridge were the most potent, with compound 11 showing a remarkable 28-fold increase in SHMT2 inhibitory activity over the previous lead analogue 1 (4-carbon bridge).Significant albeit progressively decreased inhibition of SHMT2/SHMT1 was seen with the 4-(1, 5, 14, 15, and 16) and 3-carbon bridge analogues (2, 13, and 16).Although the nature of the bridge aromatic ring (phenyl versus thiophene) had no consistent impact on enzyme target inhibition, heteroatom replacements in the bridge region were generally detrimental to inhibition of both SHMT2 and SHMT1 except for oxygen, which provided similar inhibition compared to the corresponding carbon bridge analogue.For GARFTase, both the 4-and 5-atom bridge appeared to enhance inhibition, with increased inhibition by 15 (11-fold), 3 (9.5-fold),and 9 (8.7-fold) over 1.For ATIC (AICARFTase), structural impacts on enzyme inhibition were more subtle, although a bridge thiophene (14) increased the potency ∼4-fold over 1 with a phenyl ring.
We established a broad spectrum capacity for selective FRtargeting for this series with generally limited transport for RFC and PCFT.With the exception of 1, 2, 4, 6, and 13, all compounds were inactive for RFC and/or PCFT.Interestingly, PCFT-targeting appeared to be enhanced by fluorination (compare 13 to 2 and 1 to 15).By metabolite rescue studies in KB cells, we identified mitochondrial C1 metabolism and de novo purine nucleotide biosynthesis as the targeted pathways by the pyrrolo[3,2-d]pyrimidine antifolates.By in vitro enzyme assays, we confirmed inhibition of SHMT2 by our pyrrolo[3,2d]pyrimidine inhibitors over a 1331-fold range, as well as direct inhibition of the purine biosynthetic enzymes at GARFTase and/or AICARFTase.Our discovery of novel compounds that target both mitochondrial and cytosolic C1 metabolism is particularly noteworthy in that by direct targeting mitochondrial C1 metabolism, this would exacerbate the impact of inhibiting cytosolic targets by limiting the source of C1 units for cytosolic anabolism. 21Further, inhibition of both SHMT1 as well as SHMT2 is essential as this prevents metabolic "compensation" and reversal of SHMT1 (serine → glycine) in response to loss of SHMT2. 35f course, for all of these cellular targets, extending in vitro results with monoglutamyl antifolate compounds to cells must be done with caution, reflecting the potential impact of membrane transport and polyglutamate synthesis on enzyme inhibition by antifolates in cells. 16As previously reported, 5 1 is metabolized to polyglutamate forms in both the cytosol and mitochondria of MIA PaCa2 pancreatic cancer cells.In the present study, toward KB tumor cells, 1 was the best inhibitor of cell proliferation, and 13 was somewhat more potent than 1 toward HPAC pancreatic cancer cells.These results may reflect cellular pharmacodynamic determinants of drug activity.This is reflected in the results with compound 9, considered the most promising compound of the series from SAR and enzymology studies but moderately inhibitory toward KB and HPAC tumors.SAR studies to further optimize all four targets' inhibitory potencies in a single molecule and improve KB and HPAC antitumor activity are currently underway.
In conclusion, our results document novel first-in-class conformationally flexible pyrrolo [3,2-d]pyrimidine antifolate inhibitors which provide tumor selectivity via FR selectivity and inhibition of multiple essential metabolic pathways for malignant cells including mitochondrial C1 metabolism at SHMT2 and de novo purine nucleotide biosynthesis at GARFTase and ATIC.Multitargeted inhibitors such as those described would offer substantial advantages in circumventing resistance to single-target drugs.The specificity of these agents for FRs over the ubiquitously expressed RFC, and inhibition of de novo purine nucleotide biosynthesis and mitochondrial C1 metabolism are especially notable given the association of these pathways with malignant cells. 11,13,18,19,21,74These novel multitargeted pyrrolo [3,2-d]pyrimidine agents represent an exciting new structural motif for targeted cancer therapy with substantial advantages of selectivity and potency over clinically used antifolates. 74EXPERIMENTAL SECTION General Chemistry.All evaporations were carried out at reduced pressure with a rotary evaporator.Analytical samples were dried in vacuo in a CHEM-DRY drying apparatus over P 2 O 5 at 50 °C.Melting points were determined either using a MEL-TEMP, II melting point apparatus with FLUKE 51 K/J electronic thermometer or using an MPA100 OptiMelt automated melting point system and are uncorrected.Nuclear magnetic resonance spectra for proton ( 1 H NMR) were recorded on the Bruker Avance II 400 (400 MHz) or Bruker Avance II 500 (500 MHz) NMR systems with TopSpin processing software.The chemical shift values (δ:) are expressed in, (parts per million) relative to tetramethylsilane as an internal standard: s, singlet; d, doublet; dd, doublet of doublet; t, triplet; q, quartet; m, multiplet; br, broad singlet; td, triplet of doublet; dt, doublet of triplet; quin, quintet; exch., exchangeable using D 2 O. Thin-layer chromatography (TLC) was performed on Whatman PE SIL G/UV254 flexible silica gel plates and the spots were visualized under 254 and 365 nm ultraviolet illumination.Proportions of solvents used for TLC are by volume.All analytical samples were homogeneous on TLC in at least two different solvent systems.Column chromatography was performed on silica gel (70 to 230 mesh, Fisher Scientific) column.Flash chromatography was carried out on the CombiFlash Rf systems, model COMBIFLASH RF.Pre-packed RediSep Rf normal-phase flash columns (230 to 400 mesh) of diverse sizes were used.The amount (weight) of silica gel for column chromatography was in the range of 50−100 times the amount (weight) of the crude compounds being separated.Elemental analyses were performed by Atlantic Microlab, Inc., Norcross, GA.Element compositions are within ±0.4% of the calculated values.Fractional moles of water or organic solvents frequently found in some analytical samples could not be prevented despite 24 to 48 h of drying in vacuo and were confirmed where possible by their presence in the 1 H NMR spectra.The 13 C NMR data was collected using Bruker Avance II 400 (400 MHz) NMR system with TopSpin processing software or Bruker 600 MHz Avance Neo NMR with a TCI Cryo Probe.The high-performance liquid chromatography (HPLC) was performed using UltiMate 3000 UHPLC+ system.Reversed-phase HPLC was carried out XSelect CSH C18 XP, 130Å, 2.5 μm, 3 mm × 100 mm column.Solvent A: water with 0.1% trifluoroacetic acid (TFA); Solvent B: acetonitrile.Mass spectrometry m/z determinations were performed by an Advion Expression-S CMS (a single quadrupole compact MS) controlled by Advion Chems Express 4.0.13.8 software.High-resolution mass spectroscopy (HRMS) was performed on a Thermo Scientific Q Exactive high-resolution mass spectrometer or Thermo Scientific Finnegan LTQ Orbitrap XL mass spectrometer.Purification and separation of compounds are carried out using Buchi Pure C-850 FlashPrep and XBridge Prep C18 5μM OBD 19 mm × 150 mm column.The purities of the final compounds determined by HPLC analysis were >95%.
Synthesis of 2, 3, 4, 9, 15, and 16 are already published. 36eneral Procedure for the Synthesis of 18a−b.Ethyl 4hydroxybenzoate ( 17) (1 equiv), potassium carbonate (1.5 equiv), and appropriate alcohol (1.5 equiv) were added to 200 mL of acetonitrile, and the suspension was refluxed at 95 °C for 16 h.The solvent was evaporated under reduced pressure and to the solids were added water (20 mL) and ethyl acetate (3 × 50 mL).To the ethyl acetate layer was added anhyd.sodium sulfate, and the solution was filtered.Silica gel was added to the solvent and a plug was prepared.Flash chromatography was carried out using ethyl acetate-hexane to afford 18a−b.
Ethyl 4-(2-Hydroxyethoxy)benzoate (18a).Compound 18a was synthesized using the general method described for the preparation of 18a−b using ethyl 4-hydroxybenzoate ( 17 Ethyl 4-(3-Hydroxypropoxy)benzoate (18b).Compound 18b was synthesized using the general method described for the preparation of 18a−b using ethyl 4-hydroxybenzoate (17) (6 g, 39.44 mmol), potassium carbonate (8.18 g, 59.15 mmol), and 3-bromopropanol (8.22 g, 59.15 mmol), to give 18b as a colorless oil (6.2 g, 75%); TLC R f = 0.13 (EtOAc/Hexane, 1:2); 1  General Procedure for the Synthesis of 19a−b, 23, and 31e.To the alcohols 18a−b, 22 and 30e were dissolved in dichloromethane (25 mL) and triethylamine (1 equiv) was added to the mixture.The mixture was cooled to 0 °C and stirred for 10 min.An addition funnel was attached at the top of the RBF, created vacuum inside the system, and then nitrogen gas was purged to create inert atmosphere.With the use of the addition funnel, under anhyd.conditions, methanesulfonyl chloride (1.05 equiv) was added dropwise over 30 min.The reaction was stirred at rt for 1 h and then the reaction was directly added to sodium bicarbonate solution (50 mL) in a separatory funnel and the bottom dichloromethane layer was collected.The water layer was washed with dichloromethane (3 × 100 mL).The dichloromethane was evaporated to obtain a semisolid product.To the intermediate in acetone, sodium iodide (1 equiv) was added and refluxed for 8 h.The reaction mixture was filtered.The filtrate was evaporated to obtain 19a−b, 23, and 31e.
General Procedure for the Synthesis of 30a−e.To a Parr flask were added 29a−e, 10% palladium on activated carbon (50% w/w), and MeOH (100 mL).Hydrogenation was carried out at 55 psi of H 2 for 14 h.The reaction mixture was filtered through Celite, washed with MeOH (100 mL), and concentrated under reduced pressure to give a crude mixture containing 30a−e.Without chromatographic separation, these compounds were used for the next reaction.
Methyl 3-Fluoro-4-(4-hydroxybutyl)thiophene-2-carboxylate (30e).Compound 30e was synthesized using the general method described for the preparation of 30a−e, from 29e (1.86 g, 8.4 mmol) to give 1.87 g (83%) of 30e as a pale yellow oil; TLC R f = 0.3 (EtOAc/ Hexane, 1:1); 1  General Procedure for the Synthesis of 31a−d.To a solution of triphenylphosphine (1.5 equiv) in dry methylene chloride (100 mL) were added iodine (1.5 equiv) and imidazole (1.5 equiv) at 0 °C.The resulting solution was stirred for 10 min, before a solution of 30a−d (1 equiv) in dry methylene chloride (50 mL) was added.The reaction mixture was then quenched with sat.aqueous sodium thiosulfate solution after 30 min.The organic layer was separated, washed with brine, and dried over Na 2 SO 4 .Silica gel was then added, and the solvent was evaporated under reduced pressure.The resulting plug was loaded onto a silica gel column with ethyl acetate-hexane as the eluent.Fractions with the desired peak were collected to obtain 31a−d.
Methyl 3-Fluoro-4-(4-iodobutyl)thiophene-2-carboxylate (31e).Compound 31e was prepared using the general method described for the preparation of 19a−b, 23, and 31e, from compound 30e (2.5 g, 10.40 mmol), methanesulfonyl chloride (0.96 mL, 12.34 mmol), and triethylamine (1.9 mL, 13.92 mmol) to form the intermediate.To this, sodium iodide was added and the procedure was followed to give 2.47 g (82%) of 31e as a colorless oil; TLC R f = 0.9 (EtOAc/Hexane, 1:1); 1  General Procedure for the Synthesis of 33a−e.To a solution of ethyl 3-amino-1H-pyrrole-2-carboxylate hydrochloride (0.5 g, 3.24 mmol) in dry DMF (10 mL) was slowly added NaH (0.17 g, 7.1 mmol) under nitrogen at rt.The resulting mixture was stirred for about 15 min when there was no more gas evolved, and then the appropriate iodide 31a−e (1 equiv) was added.The reaction mixture was stirred at rt for 4 h, and DMF was evaporated at elevated temp.to offer a gummy residue 32a−e, which was used for the next step without purification.The gummy residue was dissolved in MeOH (10 mL), and 1,3bis(methoxycarbonyl)-2-methyl-2-thiopseudourea (0.7 g, 3.3 mmol) was added followed by AcOH (1.0 g, 15 mmol).The mixture was stirred at rt overnight and became a thick paste.NaOMe in MeOH (25%) (7 mL, 22 mmol) was added, and stirring was continued at rt overnight.The mixture was neutralized with AcOH, and the methanol was removed under reduced pressure.To the residue was added water (20 mL), and the pH was adjusted to 10−11 by adding NH 3 •H 2 O.The solid was collected by filtration and washed well with water.The resulting solid was added to 1 N NaOH (2 mL), and the mixture was heated at 55−65 °C for 3 h.The mixture was cooled and acidified using 1 N HCl.The precipitate was collected and dried overnight under reduced pressure to obtain 33a−e.
General Procedure for the Synthesis of 34a−d.Diethyl L- glutamate HCl (1 equiv) was dissolved in DMF (20 mL).In a separate flask, pteroic acid 33a−d (1 equiv) was dissolved in DMF (20 mL).The solution of diethyl L-glutamate HCl (1 equiv) in DMF was then added, followed by DIPEA (3 equiv) and HATU (1 equiv).The mixture was stirred overnight at rt and then diluted with EtOAc (200 mL).The organic layer was washed sequentially with 50 mL of half-saturated sodium chloride, 50 mL of 10% citric acid (aq.), 50 mL of half-saturated sodium chloride, 50 mL of saturated sodium bicarbonate (aq.), 50 mL water, and then with 2 × 50 mL brine.The organic layer was dried over sodium sulfate, filtered, and concentrated under reduced pressure.Purification by column chromatography eluting 0−5% MeOH/DCM gave the desired product 34a−d.
Purification of the three enzymes was performed by Ni-NTA chromatography (Gold Biotechnology), as described for His-ATIC.After purification by Ni-NTA, pyridoxal-5′-phosphate (PLP) was added to SHMT1 or SHMT2 in 3-fold molar excess and allowed to incubate for 16−18 h.Final purification of SHMT enzymes was completed with size exclusion chromatography with a Superdex 200 16/60 column (GE Healthcare).Buffer containing 20 mM sodium phosphate, pH 7.5, 100 mM KCl, 0.2 mM EDTA, and 5 mM β-Me was used for the chromatography and selection of PLP-loaded SHMT enzyme was performed by monitoring absorbance at 435 nm.(the wavelength of the Lys-PLP covalent bond).His-MTHFD2 was bufferexchanged using a G25 Sephadex desalting column (GE Healthcare) for final purification using 50 mM Tris-HCl, pH 7.5, 250 mM NaCl, 5 mM magnesium chloride, 5% glycerol, and 0.5 mM Tris(2carboxyethyl)phosphine (TCEP) for protein assays.
Crystallization and Structure Determination of Human GARFTase and Human SHMT2.The purified human GARFTase (residues 808−1010) with a noncleavable C-terminal His 6 -tag was buffer-exchanged into 25 mM Tris-HCl, pH 8.0, 200 mM NaCl, and 0.6 mM TCEP.The protein was then concentrated to 10 mg/mL and incubated in a 3-fold ratio with α,β-GAR in excess at 4 °C for 30 min in the presence of an inhibitor.Crystals were formed by hanging drop plates containing 1 μL of protein-ligand solution, 0.8 μL of crystal condition, and 0.2 μL of 9 mM N-decyl β-D-thiomaltoside (Hampton Research, Aliso Viejo, CA), which was incubated over 0.5 mL crystallant containing 0.1 M Tris-HCl, pH 7.5, 330 mM NaCl, 16− 21% poly(ethylene glycol) (PEG) 4000, and 2% PEG 400.After a few days, cube-shaped crystals formed and were cryo-protected with a stepwise transfer into crystallant with 35% PEG 4000, then flash-frozen by immersion into liquid nitrogen. 79urified human His 6 -SHMT2 loaded with PLP was crystallized in bulk solution.The protein was stored in buffer containing 20 mM sodium phosphate, pH 7.5, 100 mM NaCl, 0.2 mM EDTA, and 0.5 mM TCEP at 10 to 20 μM.After storing at 4 °C for 5−7 days, rod-shaped crystals formed.Crystals were soaked with serine and inhibitor at a 5:5:1 ratio of serine:inhibitor:His-SHMT2 and transferred to crystallant containing 35% glycerol, 100 μM serine, and 100 μM inhibitor for cryo-protection before being plunged into liquid nitrogen.
Data collection was performed at the Advanced Light Source beamline 4.2.2 using the Taurus CMOS detector at Lawrence Berkeley National Laboratory.Data sets for GARFTase were processed in space group P3 2 2 (XDS). 80,81Protein Data Bank (PDB) entry 1J9F with waters and ligands removed was used as the molecular replacement search model (PHE-NIX). 82Data sets for SHMT2 were processed to space group P6 5 22 (XDS).Molecular replacement was performed using PDB ID: 5V7I with waters and ligands removed as the search model. 83oot was used for model building and Phenix was used for structure refinement. 82,84n Vitro Enzymatic Assays and K i Determinations.Inhibition of GARFTase activity was evaluated by monitoring the conversion of 10formyl THF to THF by absorbance at 298 nm in the presence of an inhibitor. 36,48The GARFTase inhibition assays utilized 40 μM 10formyl THF, 50 nM His-GARFTase, 15 μM α,β-GAR, and a range of inhibitor concentrations (final concentrations) in assay buffer containing 25 mM Tris-HCl, pH 8, 300 mM NaCl, and 5 mM β-Me at 37 °C. 36AICARFTase activity was also measured by monitoring the conversion of 10-formyl THF to THF by absorbance at 298 nm in the presence of inhibitor. 36The AICARFTase inhibition assays utilized 50 μM 10-formyl THF, 100 nM His-ATIC, 50 μM ZMP, and a range of inhibitor concentrations in assay buffer containing 32.6 mM Tris-HCl, pH 7.5, 25 mM KCl, and 5 mM β-Me at 25 °C. 36MTHFD2 activities were assayed by a reaction with methylene THF and NAD + .NADH production was measured by fluorescence (excitation 340 nm, emission 480 nm).Reaction conditions included 100 nM MTHFD2, 5 μM methylene THF, and concentrations of small-molecule inhibitor ranging from 0.02 nM to 2000 nM; NAD + (2.5 mM) was added to initiate the reaction.The reaction buffer consisted of 50 mM Tris-HCl, pH 7.5, 250 mM NaCl, 5 mM magnesium chloride, 5% glycerol, and 0.5 mM TCEP.Kinetic measurements were recorded in triplicate using a flat, black-bottom, black-walled 96-well plate (Greiner Bio-One, 655900) and a BioTek Synergy Neo2 Plate Reader.MTHFD2 Inhibition was not detected with any of the small-molecule inhibitors in this work.To assay SHMT1 or SHMT2 (SHMT1/2) enzymes, a coupled reaction with His-MTHFD2 in 200-fold molar excess was used where the production of NADH was monitored by fluorescence with excitation at 360 nm and emission at 470 nm at 25 °C.The reactions contained 5 nM His-SHMT1/2, 50 μM THF, 20 mM serine, 10 μM His-MTHFD2, 2.5 mM NAD + , and a range of inhibitor concentrations (final concentrations).The reactions were performed in buffer containing 20 mM sodium phosphate, pH 7.5, 100 mM KCl, 0.2 mM EDTA, and 5 mM β-Me.A BioTek Synergy Neo2 Plate Reader was used to record kinetic measurements in triplicate in a UV-transparent 96-well plate (Costar 3635) for GARFTase and AICARFTase assays and a black-well, black-bottom 96-well plate (Corning 3603).To calculate IC 50 values for each inhibitor, initial velocities were graphed as a function of inhibitor concentrations and fit with a three-parameter nonlinear regression (GraphPad Prism 8.0).The equation [K i = IC 50 / ([S]/K M + 1)] was used to convert IC 50 's to K i values.K M values were calculated as 84.8 μM for His-GARFTase, 100 μM for His-ATIC, 63 μM for SHMT1, and 108 μM for SHMT2. 36tatistics.All data reflect at least three biological replicates unless noted otherwise.All statistical comparisons were performed using unpaired t-tests after data were transformed to meet normality assumptions, and no p-value adjustments were made for multiple comparisons.A log 2 transformation was used for data with positive values and, when data included zero values, a square root transformation was used.For depicting data in plots, all data were summarized with mean values and standard deviations using data without transformation.

Data Availability Statement
Atomic coordinates and structure factors for the reported crystal structures have been deposited in the Protein Data Bank (8FJT, 8FJU, 8FJV, 8FJW, 8FJX, 8FJY); see Table S2, Supporting Information.The authors will release the atomic coordinates and experimental data upon article publication.

Figure 2 .
Figure 2. Structures of the clinically used classical antifolates.
The C−X bond lengths of these compounds follow the trend C−O < C−NH < C−CH 2 < C−S (Figure 5).The C−X−C bond angle follows the trend C−S−C < C−CH 2 −C < C−NH−C < C−O−C.Thus, replacing the benzylic CH 2 can vary the bond angle and/or bond length (Table

Figure 6 .
Figure 6.Structural analysis of 5-substituted pyrrolo[3,2-d]pyrimidine 1 in SHMT2-PLP-gly/PLG-antifolate ternary complexes.(A) Crystal structures of 1 (orange) bound in the THF binding pocket of both pockets of the asymmetric unit of the SHMT2 dimer with the natural cofactor PLP (purple).1 occurs in a single conformation (conformation A) in pocket A. (B) Crystal structures of 1 (orange) with the natural cofactor PLP (purple) in two conformations (orange and teal) in pocket B (conformation A had an occupancy of 41% and conformation B had an occupancy of 59%).

Figure 7 .
Figure 7. Superposition of 1 in each binding pocket of crystallized SHMT2.Superposition of crystallized 1 in pocket A (orange) and pocket B (dark gray) in SHMT2.

Figure 8 .
Figure 8. Conformations of 1 in SHMT2 structure.Comparison of conformations of 1 in molecular modeling and X-ray crystal structures.(A) Energyminimized structure of 1.The distance between F−H−N− is 4.35 Å.The energy of this conformation is −260.6 kcal/mol.(B) Bound crystal structure of 1 in SHMT2 pocket A. The distance between F−H−N is 4.69 Å.The energy of this conformation is −243.8kcal/mol.(C) Bound crystal structure of the alternate conformation of 1 in SHMT2 pocket B with the distance between F−H−N at 1.84 Å.The energy of this bound conformation is −258.8kcal/mol.

Figure 9 .
Figure 9. Crystal structures of inhibitors bound in the folate-binding pocket of GARFTase.Crystal structures were determined of GARFTase in complex with 3 (A, PDB ID: 8FJX), 1 (B, PDB ID: 8FJW), or 2 (C, PDB ID: 8FJY) (blue) and substrate GAR (tan).The protein is represented as a gray ribbon with interacting residues shown as sticks.

Scheme 2 a 3 a 4 a
Scheme 2

Figure 11 .
Figure 11.Structural analysis of 5-substituted pyrrolo[3,2-d]pyrimidine analogue 14 in complex with SHMT2-PLP-gly-antifolate ternary complexes.(A) Crystal structure of 14 bound in the THF binding pocket A of the asymmetric unit of the SHMT2 dimer including the natural cofactor PLP bound to glycine (PLG) (gold).(B) Crystal structure of 14 bound in the THF binding pocket B of the asymmetric unit of the SHMT2 dimer including the natural cofactor PLP bound to Lys280 (LLP) (salmon).

Figure 12 .
Figure 12.Comparison of 14 conformations in the molecular modeling and X-ray crystal structures for human SHMT2.(A) Energy-minimized structure of 14.The distance between F−H−N is 4.26 Å, the distance between F−O is 3.13 Å, and the distance between S−H−N is 2.82 Å. (B) Ligandbound crystal structure of 14 in SHMT2 pocket A with the distance between F−H−N of 2.61 Å; the distance between F−O is 4.03 Å; and the distance between S−H−N is 3.94 Å. (C) Ligand-bound crystal structure of 14 in SHMT2 pocket B with the distance between F−H−N of 2.12 Å; the distance between F−O is 4.21 Å; and the distance between S−H−N is 4.08 Å.

Table 2 .
K i 's for Inhibition of Enzymes for 5-Substituted Pyrrolo[3,2-d]pyrimidine Antifolates a Journal of Medicinal Chemistry docked structure conformation along with the inhibitory potency.
demonstrating tumor cell selectivity over normal cells.Identification of the Targeted Pathway(s) in KB Tumor Cells by Metabolite Rescue.To confirm the targeted pathways for the pyrrolo[3,2-d]pyrimidine antifolates in tumor cells, we

Table 1 .
Distances and Bond Angle Variations for 4 Atom Bridge Pyrrolo[3,2-d]pyrimidine Antifolates Predicted by the Atom of the Bridge at the Benzylic Position (X) a Distances and angles for X = NH were measured using energyminimized conformations of compounds with Maestro12.1.(Maestro, Schrodinger, LLC, New York, NY, 2021).
45Bond angles for X = CH 2 , O, and S obtained from the literature.45

Table 2 . continued a
For the in vitro enzyme assays, K i values are presented as mean values (±standard deviations) from at least 3 replicate experiments.Methods are described in the Experimental Section.

Table 3 .
IC 50 's for Inhibition of Proliferation of Isogeneic Chinese Hamster Ovary Cells and Human Tumor Cells by 5-Substituted Pyrrolo[3,2-d]pyrimidine Antifolates a