Serine hydroxymethyltransferase as a potential target of antibacterial agents acting synergistically with one-carbon metabolism-related inhibitors

Serine hydroxymethyltransferase (SHMT) produces 5,10-methylenetetrahydrofolate (CH2-THF) from tetrahydrofolate with serine to glycine conversion. SHMT is a potential drug target in parasites, viruses and cancer. (+)-SHIN-1 was developed as a human SHMT inhibitor for cancer therapy. However, the potential of SHMT as an antibacterial target is unknown. Here, we show that (+)-SHIN-1 bacteriostatically inhibits the growth of Enterococcus faecium at a 50% effective concentration of 10–11 M and synergistically enhances the antibacterial activities of several nucleoside analogues. Our results, including crystal structure analysis, indicate that (+)-SHIN-1 binds tightly to E. faecium SHMT (efmSHMT). Two variable loops in SHMT are crucial for inhibitor binding, and serine binding to efmSHMT enhances the affinity of (+)-SHIN-1 by stabilising the loop structure of efmSHMT. The findings highlight the potency of SHMT as an antibacterial target and the possibility of developing SHMT inhibitors for treating bacterial, viral and parasitic infections and cancer.

O ne-carbon (1C) metabolism is an essential pathway for constructing fundamental biomolecules necessary for prokaryotic and eukaryotic cell growth and proliferation [1][2][3][4] . 1C metabolism encompasses a complex metabolic network involving folate-dependent chemical reactions. In 1C metabolism, folate compounds are cyclically used or recycled by folate-dependent enzymes, such as dihydrofolate reductase (DHFR), serine hydroxymethyltransferase (SHMT) and thymidylate synthase (TS). DHFR converts dihydrofolate to tetrahydrofolate (THF). SHMT produces 5,10-methylenetetrahydrofolate (CH 2 -THF) from THF by converting serine into glycine. CH 2 -THF acts as a 1C unit carrier and is used as a fundamental building block for essential biomolecules (e.g., amino acids and nucleotides) 2,4 . Glycine is used for purine synthesis in several pathways downstream of 1C metabolism. In addition, TS converts deoxyuridine monophosphate to thymidine monophosphate (dTMP) by transferring the 1C unit from CH 2 -THF. Subsequently, dTMP is phosphorylated to form thymidine triphosphate, a DNA nucleotide.
Suppressing 1C metabolism using chemicals is effective in cancer and antibacterial therapies. Pemetrexed (PMX) and methotrexate (MTX) are folate antagonists in human cells and inhibit DHFR in cancer cells 5 , whereas 5-fluorouracil (5-FU) targets nucleotide metabolism, including thymidine synthesis 6 . These compounds are major clinical anticancer agents that have been used as frontline chemotherapies to treat various cancers. Trimethoprim (TMP) is used as a bacterial DHFR inhibitor in antibacterial therapy. We reported previously that 5-fluoro-2′deoxyuridine (5-FdU) inhibits bacterial TS and induces thymineless death in Staphylococcus aureus 6 .
SHMT, a pyridoxal 5′-phosphate (PLP)-dependent enzyme, catalyses the retro-aldol cleavage of serine to glycine and converts THF into CH 2 -THF 7,8 . The glycine and CH 2 -THF supplied by SHMT are essential for pyrimidine, purine and amino acid syntheses 9,10 . Therefore, drugs targeting SHMT can block multiple metabolic pathways downstream of 1C metabolism (e.g., DNA and RNA synthesis), indicating that SHMT is a potential target of therapeutic agents directed at various viruses, bacteria and parasite infections, as well as cancer. Indeed, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has been shown to induce purine synthesis to support the synthesis of viral subgenomic RNA by hijacking serine metabolism through human SHMT1 11 . Moreover, SHMT inhibitors under development are promising therapeutic agents against Plasmodium parasites 12,13 . Furthermore, novel chemical tools for cancer chemotherapy have been developed that inhibit SHMT activity by targeting the production of 1C units from serine, a primary 1C unit source in cancer cells 14 . The pyrazolopyran derivatives (+)-SHIN-1 and SHMT-IN-2 have anticancer activities against large, diffuse B-cell lymphoma by inhibiting human SHMT1/2 activity 14 . Sertraline (SER), a neuromodulatory drug, has also been found to suppress breast tumour growth by inhibiting SHMT1/2 15,16 .
In contrast to cancer treatment, the effect of SHMT inhibition on prokaryotic cells is unclear. The Enterococcus genus requires folate-related compounds for growth and incorporates these compounds 17 . In particular, Enterococcus faecium has been used for studying the bacterial folate cycle and for drug screening because this bacterium grows in folate-free media that includes amino acids and nucleic acids (purine and pyrimidine bases) 18 . E. faecium is susceptible to bacterial TS-inhibiting pyrimidine analogues such as 5-FdU 19 . Thus, E. faecium is a suitable bacterium for determining the potential of SHMT as an antibacterial target and for investigating the synergistic effects of inhibiting 1C metabolism using a combination of DHFR, TS and SHMT inhibitors.
Here, we investigated SHMT as an antibacterial target using E. faecium and (+)-SHIN-1. (+)-SHIN-1 activity was highly potent against E. faecium with 50% effective concentration (EC 50 ) values on the order of 10 -11 M. We found that excess deoxythymidine (dT) reduced (+)-SHIN-1 activity, suggesting that the major antibacterial mechanism involved inhibiting physiological thymidine synthesis by blocking the supply of the 1C-unit carrier of 1C metabolism, CH 2 -THF. Differential scanning fluorimetry and crystal structural analyses revealed that (+)-SHIN-1 acts as an E. faecium SHMT (efmSHMT) inhibitor and that the biphenyl and hydroxy groups in (+)-SHIN-1 are important for high-affinity binding to efmSHMT. E. faecium is also susceptible to DHFR inhibitors, such as PMX, MTX and raltitrexed (RTX), and 5-FdU, which is also reported as a TS inhibitor 19 . Thus, E. faecium is susceptible to compounds inhibiting three major enzyme types in 1C metabolism. We found that (+)-SHIN-1 enhances the antibacterial activity of nucleoside analogues, suggesting that it has a synergistic effect when used with DNA and RNA synthesis inhibitors. These results shed light on the potential of SHMT as an antibacterial target and support the development of SHMT inhibitors as therapeutic agents for bacterial, viral and parasite infections and cancer.

Results
Bacterial sensitivity to 1C metabolism inhibitors. The potency of SHMT as an antibacterial target was evaluated by determining the antibacterial activities of several 1C metabolism inhibitors on E. faecium, E. faecalis, S. aureus and E. coli (Fig. 1a Determination of time-dependent bacterial viability with or without inhibitors. We evaluated the time-dependent change in bacterial viability in the presence and absence of each inhibitor to quantitate the bacterial cell growth carefully and determine the inhibition mechanism of the compounds (Fig. 2 Antibacterial assay with excess concentrations of serine, glycine and physiological nucleosides. The antibacterial mechanism of (+)-SHIN-1 and other compounds was determined by performing antibacterial assays with excess concentrations of folic  acid, Ser, Gly, inosine and dT when compared with the physiological concentrations of these molecules (Fig. 3a). In the presence of 100 mM Ser, the antibacterial activities of (+)-SHIN-1 and SHIN-1 were 10-and 5-fold weaker than when Ser was absent ( Table 2). The bacterial viability assay showed that E. faecium growth was affected in the presence of 100 mM Ser (Supplementary Fig. S1) and that the antibacterial activity of (+)-SHIN-1 with 100 mM Ser was weaker when compared with the antibacterial activity of (+)-SHIN-1 in the absence of Ser. The decrease in antibacterial activity with Ser was not caused by competition between (+)-SHIN-1 and Ser. PMX also displayed weaker antibacterial activity in the presence of 100 mM Ser. The other compounds, including SHMT-IN-2 and SER, showed similar activities with/without 100 mM Ser. In the presence of 10 mM Gly, (+)-SHIN-1 and SHIN-1 displayed 3-and 2-fold weaker activities than when Gly was absent. The activity of 5-fluorouridine (5-FUrd) was slightly stronger in the presence versus absence of Ser and Gly. In the presence of 50 and 100 mM Gly, the antibacterial activity of each drug was not detectable because Gly caused cell toxicity ( Supplementary Fig. S2). In the presence of 1 mM inosine, compounds except (+)-SHIN-1 and 5-FUrd displayed similar activities to those when inosine was absent. As shown in Fig. 3a, an excess amount of Ser, which is an SHMT substrate, can counteract the activity of (+)-SHIN-1. Abundant inosine levels potentially offset the nucleotide imbalance caused by decreased levels of purine bases. The antibacterial activities of 5-FUrd and (+)-SHIN-1 with 1 mM inosine were weaker (4-and 13-fold, respectively) than those without inosine.
In the presence of 1 mM dT, tested compounds except SHMT-IN-2 and SER lost their antibacterial activities. This observation suggests that 1 C metabolism inhibitors cause thymine starvation in E. faecium, but the antibacterial mechanism of SHMT-IN-2 and SER differs from these compounds 20,21 . The antibacterial activity of (+)-SHIN-1, SHIN-1 and folate analogues decreased when 0.1 mM folate (FA) was present, whereas the antibacterial activity of SHMT-IN-2 and SER was independent of FA ( Table 2). Thus, SHMT-IN-2 and SER do not target folate cyclerelated metabolic enzymes, including SHMT.
Thermal stability of efmSHMT with or without each compound. We examined the interaction between efmSHMT and each compound using differential scanning fluorimetry (DSF) 22,23 . As illustrated in Fig. 3, the thermal stability of efmSHMT only shifted in the presence of (+)-SHIN-1 (ΔT m = 7°C), indicating that (+)-SHIN-1 bound to efmSHMT. SHMT-IN-2, SER and MTX did not stabilise efmSHMT significantly, whereas PMX (ΔT m = 0.9°C) stabilised efmSHMT slightly ( Supplementary Fig. S3). These data support the results of the competition assays. Therefore, (+)-SHIN-1 is an efmSHMT inhibitor and SHMT-IN-2, SER and MTX are not. PMX binds to efmSHMT, but the affinity is probably quite weak. The DSF results using ecSHMT and each compound indicated that (+)-SHIN-1 bound to ecSHMT ( Supplementary Fig. S4), but other compounds did not. Antibacterial assays and DSF results suggested that (+)-SHIN-1 was not captured by E. coli. The biphasic nature of the melting curve induced by (+)-SHIN-1 complicated melting point calculations. This biphasic curve may have arisen because of monomer/dimer melting.
Evaluation of the binding affinities of each compound against efmSHMT. The binding affinities of each compound against efmSHMT were evaluated using a binding assay that utilised the 500nm absorption value derived from quinonoid formation in the SHMT ternary complex of SHMT, Gly and 5-methyltetrahydrofolate (Me-THF) (efmSHMT/Gly/Me-THF) 24,25 . In the presence of excess Gly, the efmSHMT/Gly/Me-THF complex forms quinonoid, whereas the efmSHMT-inhibitor complex does not. Therefore, adding a potential inhibitor to the ternary complex decreases the absorbance at 500 nm (OD 500 ) derived from quinonoid if the compound functions as an SHMT inhibitor (Fig. 4a). Using this feature and the Cheng-Prusoff equation, we estimated the K i of compounds.
detection limit concentration of efmSHMT was 2.5 μM, and the concentration was too high to evaluate the affinity of (+)-SHIN-1 accurately. In contrast, the lowest SHMT concentration used represents the lower limit K i of (+)-SHIN-1 determined in this assay. Therefore, the actual K i is probably lower than 0.0088 μM, indicating the extremely strong affinity of (+)-SHIN-1 toward efmSHMT. The DSF and the binding assay results showed that the main target of (+)-SHIN-1 is SHMT in E. faecium, which is the same as in humans. Furthermore, the binding affinity of (+)-SHIN-1 toward efmSHMT was enhanced in the presence of Ser ( Supplementary  Fig. S6).
Structural comparisons between efmSHMT and three reported SHMTs. We solved the structures of efmSHMT in complex with pyridoxal 5ʹ-phosphate (PLP) and (+)-SHIN-1 in the absence or presence of Ser ( Fig. 5 and Table 3) at 2.28 and 1.90 Å resolutions, respectively, to characterise SHIN-1 binding to efmSHMT. In addition to these structures, the efmSHMT/Gly/Me-THF complex structure was determined at 2.62 Å resolution. DLS measurements indicated that the hydrodynamic radius of efmSHMT was 4.209 nm with an evaluated molecular weight of 97.2 kDa (theoretical monomer molecular weight is 44.7 kDa), suggesting efmSHMT forms a dimer in solution. In the crystal structures, we observed that the asymmetric unit includes two molecules that form a dimer.
In the complex structure with (+)-SHIN-1, electron density was observed in both binding pockets of the protein dimer ( Fig. 5b), and the (+)-SHIN-1 binding sites were well resolved. The C4′ carbon in the PLP ring was covalently linked as a Schiff's base to K226, as previously reported for the huSHMT structure 14 .
The solved structure showed that (+)-SHIN-1 binds to the folatebinding site of efmSHMT. The structure of the binding site of efmSHMT was compared with previously reported structures of human, E. coli and Plasmodium vivax SHMTs, huSHMT2, ecSHMT and pvSHMT, respectively, by superimposing the efmSHMT catalytic site on each SHMT structure (PDB IDs: 5V7I, 1DFO and 5GVN, respectively) (Fig. 6). The backbone root-mean-square deviations of the superimposed structures were 1.36 Å for huSHMT2, 0.97 Å for ecSHMT and 1.35 Å for pvSHMT. In the catalytic site, we found that two variable loop regions encompassing amino acids (aa) 115-137 (115-loop) and 343-357 (343-loop) in efmSHMT differed from each other (Fig. 6). The aa sequences of the 115-and 343-loops in efmSHMT differ markedly from those in huSHMT2 and pvSHMT, with an aa homology of <40% (Supplementary Table S4). In contrast, the 115-loop in ecSHMT has a similar structure to that found in efmSHMT. The 343-loop structure of efmSHMT differs from those of the other enzymes. The homologies of the 115-and 343loops between efmSHMT and ecSHMT were 74% and 60%, respectively (Supplementary Table S3). As previously reported, we found that the 343-loop is flexible, and its position depends on the compounds that bind to the catalytic site 13,26 . The other loops in the catalytic site are inflexible and occupy conserved positions 13 . Therefore, the difference in the 343-loop structure between efmSHMT and ecSHMT is possibly related to substrate binding to each SHMT, whereas the differences in the 115-loop structures are associated with aa sequence differences.
Ligand interactions among efmSHMT complexes. We identified the following interactions between the efmSHMT and (+)-SHIN-1 complex. The exocyclic amine in (+)-SHIN-1 forms hydrogen bonds with the main chain of Leu117 and Gly121 (Fig. 5c). This exocyclic amine also interacts with Asn343 and Ser344 through a water molecule. The pyrazole moiety forms a hydrogen bond and van der Waals interactions with the main chain of Leu123 and  The 50% effective concentration (EC 50 ) represents the range determined from at least three independent experiments. pEC 50 is log-transformed EC 50 . All data represent the mean ± standard deviation (n = 3 and n = 6 for (+)-SHIN-1). aFold increase is the ratio of the EC 50 value with an excess amount of supplement and the EC 50 value without any supplement against E. faecium, as described in Table 1. The E. faecium strain is JCM5804.
In the structure of the efmSHMT/Ser/(+)-SHIN-1 complex, the covalent bond formed by a Schiff's base relocated from K226 to the amino group of Ser (Fig. 5e). The Ser amino acid forms hydrogen bonds with E53' (prime denotes the neighbour molecule) and H122 to stabilise the 115-loop, and the carbonyl group of Ser also forms a salt bridge with R359. Similarly, the structure of efmSHMT/Gly/Me-THF showed the Schiff's base between PLP and Gly (Fig. 5f). As mentioned above, ecSHMT and efmSHMT share structural similarity. The binding mode of folate (Me-THF in efmSHMT and formyl-THF in ecSHMT) is also conserved. In the structure of efmSHMT/Gly/Me-THF, Me-THF forms hydrogen bonds with the side chains of E53' and N343 and main chain amide groups of L117, G121 and L123 (Fig. 5f).
Quantitation of mRNA for folate-binding proteins. To clarify the folate cycle of E. faecium, we identified folate-binding proteins coded in the E. faecium genome (Genbank code: UFYJ01000001.1) and likely to be part of the folate cycle of E. faecium (Fig. 3a). As a result, E. faecium uses SHMT as a primary pathway to supply 1 C units and the glycine cleavage system (GCS) as a support pathway to produce CH 2 -THF. Additionally, E. faecium was found to have a formyltetrahydrofolate synthetase (Fhs) 27 , providing CH 2 -THF by mediating FolD (Fig. 3a). The mRNA levels of the identified folate-binding proteins in the presence and absence of (+)-SHIN-1 were compared by realtime PCR (RT-PCR) to analyse the antibacterial mechanism of (+)-SHIN-1. The mRNA levels of SHMT and glycine cleavage system H protein (GcvH) were found to increase slightly 4 h after adding (+)-SHIN-1 to E. faecium cells in the growth phase (Fig. 2a). GCS also supplies CH 2 -THF from THF and Gly. Fhs and cobalamin-independent methionine synthase (MetE) mRNA levels decreased slightly, whereas the mRNA of other folatebinding proteins did not change (Fig. 3b). The mRNA levels of folate-binding proteins, except TS and DHFR, increased 10 h after adding (+)-SHIN-1 to E. faecium cells in the stationary phase (Fig. 2a, Fig. 3b). In particular, the mRNA amount for the expression of TS decreased drastically. Thus, thymidine starvation occurred in E. faecium at this time point.
Synergistic effects among 1C metabolism inhibitors including (+)-SHIN-1. To investigate whether 5-FdU and (+)-SHIN-1 have synergistic antibacterial effects, we determined the EC 50 for these two inhibitors using various drug combinations and plotted isobolograms (Fig. 7) 28 . Data points falling within the lower lefthand side of an isobologram indicate a synergism effect. As shown in Fig. 7 and Supplementary Tables S4 to S6, (+)-SHIN-1 synergistically enhanced the antibacterial activity of 5-FdU and 5-FUrd. This observation indicates that the SHMT inhibitor synergistically enhanced the antibacterial activities of these compounds, which then inhibited DNA and RNA synthesis downstream of 1C metabolism. MTX also synergistically enhanced the antibacterial activity of 5-FdU and 5-FUrd. Additionally, combined (+)-SHIN-1 and MTX acted synergistically in bacterial cells. The synergistic effect of SHMT inhibition with MTX in T-cell acute lymphoblastic leukaemia has been reported 29 .

Discussion
In this report, we determined that (+)-SHIN-1 functions as an SHMT inhibitor against E. faecium with extremely high potency and low cell cytotoxicity. (+)-SHIN-1 also exhibited synergism with 5-FdU and 5-FUrd. As previously reported, we found that 5-FdU blocks thymidine synthesis by competitively inhibiting deoxyuridine phosphatase and TS. 5-FUrd inhibited RNA synthesis as a uridine-kinase competitor. (+)-SHIN-1 inhibited SHMT and blocked the production of CH 2 -THF, which is an essential biomolecule for thymidine and purine synthesis. Thus, in combination, (+)-SHIN-1 and these nucleoside analogues downregulated several pathways related to pyrimidine and purine synthesis, thereby synergistically enhancing thymidine starvation and creating a nucleotide imbalance. Additionally, a single dose of (+)-SHIN-1 acted bacteriostatically and not as a bactericide. E. faecium utilises the supportive GCS and Fhs-FolD pathways to produce THF and CH 2 -THF (Fig. 3a). The data suggested that (+)-SHIN-1 exerts a bacteriostatic rather than a bactericidal effect because E. faecium can recycle THF and CH 2 -THF and prevent cell death by regulating the expression levels of folatebinding proteins, including GcvH and Fhs, as mRNA quantitation experiments showed. Previously, Alfadhli et al. reported that E. coli mutants that lack SHMT were unable to synthesise glycine and failed to proliferate in the absence of glycine 30 . In a previous report, E. coli lacking SHMT was able to grow in the presence of excess Gly because CH 2 -THF was supplied through GCS in the presence of excess Gly and also guanosine triphosphate (GTP) was provided through the purine synthesis pathway. In contrast, we showed that E. faecium was unable to grow in the presence of excess Gly. These observations revealed differences in the folate synthesis pathway between E. coli and E. faecium. Thus, E. coli synthesised folate using GTP, but E. faecium lacked the pathway to provide GTP. Therefore, E. faecium cannot synthesise folate even in the presence of excess Gly, suggesting that E. faecium growth in the absence of folate is hampered. Moreover, (+)-SHIN-1 may inhibit SHMT and GCS-related proteins; however, the results from the competitive binding assay indicated that the main target of (+)-SHIN-1 is efmSHMT. Further research is required to clarify these possibilities.
The potent activity of (+)-SHIN-1 was only directed against E. faecium among the tested bacteria. Enterococcus spp. and Lactobacillus spp. require folate-related compounds for growth, and both can incorporate them 18 . The other microorganisms cannot  incorporate folate derivatives. Therefore, (+)-SHIN-1 would not be incorporated by other bacteria, as observed with DHFR inhibitors, suggesting that further improvements in the chemical structure of (+)-SHIN-1 are required for developing novel antibiotics that exhibit potent activities against a broad range of bacteria. Nevertheless, as shown using the competitive assay, the antibacterial mechanism underlying SHMT inhibition involves thymine and, in part, purine starvation. These antibacterial mechanisms share similarities with TMP, which exerts potent activities against Gram-positive and -negative bacteria. Furthermore, a previous report showed that SHIN-1 lacked pharmacokinetic properties suitable for the in vivo study of SHMT biology, but a SHIN-1 derivative, which acted as a human SHMT inhibitor, overcame these pharmacokinetic issues and exerted effective anticancer activity in vivo 29 . Therefore, SHMT inhibition is a potential target of novel antibiotics. The crystal structure of the SHMT/(+)-SHIN-1 complex revealed that (+)-SHIN-1 binds strongly to efmSHIMT by forming several hydrogen bonds and hydrophobic interactions. The pyrazolopyran core of (+)-SHIN-1 interacts with the catalytic site of efmSHMT by forming hydrogen bonds with the main chain. A hydrogen bond formed with the main chain of Leu123, located in the 115-loop, is highly conserved among SHMTs. These hydrogen bonds should contribute to the high genetic barrier of (+)-SHIN-1. Darunavir, a protease inhibitor used for human immunodeficiency virus (HIV) infectious diseases, also forms strong hydrogen bonds with mainchain residues in the catalytic site of the HIV protease 23,31 . These hydrogen bonds contribute to the high genetic barrier of Darunavir 32 . Indeed, in vitro drug-resistance induction suggests that Darunavir prevents HIV-1 protease developing significant resistance using in vitro drug-resistance induction 33 . Thus, the pyrazolopyran core of (+)-SHIN-1 should prevent efmSHMT developing resistance by forming hydrogen bonds with the main chain of efmSHMT. Furthermore, the competitive binding assay showed that the binding affinity of (+)-SHIN-1 to efmSHMT was enhanced under coexistence with Ser when compared with Gly, while that of Me-THF did not change. In the structure of the efmSHMT/Ser/(+)-SHIN-1 ternary complex, the serine residue on PLP formed hydrogen bonds with E53 and H122. H122 is located on the 115-loop, and this hydrogen bond stabilises the loop. Structural analysis revealed that (+)-SHIN-1 forms several hydrogen bonds with amino acids in the 115-loop. Loop stabilisation contributes to the binding affinity of (+)-SHIN-1 to efmSHMT. In contrast, the binding affinity of Me-THF was not altered in the assay, as mentioned above. Me-THF also hydrogen bonds with amino acids in the 115-loop. Loop stabilisation can also contribute to the binding affinity of Me-THF. In the efmSHMT/Gly/Me-THF structure, E53 formed a hydrogen bond with Me-THF. Although the loss of the hydrogen bond with the 115-loop may negatively affect the binding affinity of Me-THF, the hydrogen bond with N10 positively contributes to the binding affinity. For Me-THF binding, each contribution was countered by hydrogen bond switching, and no change in the binding affinity was observed. Notably, we found that in E. faecium, (+)-SHIN-1 and SHMT-IN-2 exhibit quite distinct antibacterial mechanisms, and SHMT-IN-2 did not bind to efmSHMT despite both compounds having the same pyrazolopyran core. These two compounds have different substructures at the biphenyl moiety position in (+)-SHIN-1. Thus, the chemical structure at this position is clearly important for the specificity of SHMT. We found that the biphenyl moiety in (+)-SHIN-1 mainly interacts hydrophobically with the aa side chains in the 115-and 343loops. The exocyclic hydroxy group in (+)-SHIN-1 clearly interacts with the 342-loop. As reported for pvSHMT, the structure of the 343-loop is highly flexible. Additionally, the 115-and 343-loops share low sequence homology. The biphenyl and hydroxy groups in (+)-SHIN-1 stabilised efmSHMT/(+)-SHIN-1 binding by reducing the flexibility of the 343-loop. However, SHMT-IN-2 neither formed a hydrogen bond with the 343-loop nor bound to efmSHMT. Furthermore, Ser binding to efmSHMT enhanced the binding affinity of (+)-SHIN-1 by stabilising the 115-loop. However, there were no direct interactions between (+)-SHIN-1 and Ser complexed with efmSHMT. Thus, for the biphenyl moiety in (+)-SHIN-1, developing modifications that maintain the interaction with the 343-loop should generate novel, effective SHMT inhibitors that display species specificity. The modification using SHMT-IN-2 as a basic structure would facilitate the generation of huSHMT specific anticancer drugs, which do not affect intestinal flora. Furthermore, developing the pyrazolopyran moiety for direct interaction with Ser complexed with efmSHMT should yield more potent SHMT inhibitors than (+)-SHIN-1.
It was previously reported that PMX could bind to human SHMT-2 and Arabidopsis thaliana SHMT 25,34 . Thus, PMX could work as a species-wide SHMT inhibitor. This information is also useful to develop novel more potent and specific inhibitors of the emerging anti-bacterial, -cancer and -parasite drug target. On the other hand, in this study, it was proved that PMX bound to efmSHMT although the binding affinity was too weak to work as an efmSHMT inhibitor within the concentration range exerting antibacterial activity. Additionally, SHMT-IN-2 and SER, which are reported to be huSHMT inhibitors, never suppress SHMT activity as competitive binding assay and DSF proved. SER reportedly works as an efflux inhibitor against E. coli 35 and probably against E. faecium. SHMT-IN-2 also failed to bind to SHMT and did not compete against any competitor we tested. This observation suggests that SHMT-IN-2 will likely fail to inhibit SHMT and other enzymes involved in folate-mediated 1 C metabolism. Further research is required to determine the detailed mechanism underlying the antibacterial activity of SHMT-IN-2.
In our results, the SHMT inhibitor synergistically enhanced the antibacterial activities of nucleoside analogues. Hydroxyurea, which decreases purine triphosphate levels by inhibiting ribonucleotide reductase, reportedly enhances the type 1 anti-human immunodeficiency virus activity of azido-3ʹ-deoxythymidine, 2ʹ,3ʹ-dideoxycytidine and, in particular, 2ʹ,3ʹ-dideoxyinosine [36][37][38] . SARS-CoV-2 is reported to induce purine synthesis, supporting massive viral subgenomic RNA by hijacking serine metabolism through human SHMT1 11 . Therefore, SHMT inhibitors might enhance the anti-SARS-CoV-2 activities of RNA-dependent RNA polymerase inhibitors such as remdesivir. Thus, SHMT inhibitors may also play important roles in combination therapies for bacterial and viral infections.
In conclusion, we have shown that SHMT is a novel antibacterial target in E. faecium. New antibiotic classes are urgently required to develop therapeutics that possess potent activities against multi-drug resistant bacteria and cannot induce crossresistance to currently approved antibiotics. We found that (+)-SHIN-1 exerts strong antibacterial activity and combining (+)-SHIN-1 and nucleoside analogues produced strong synergistic effects. The crystal structure of the efmSHMT/(+)-SHIN-1 complex showed that the pyrazolopyran core of (+)-SHIN-1 forms several hydrogen bonds with the main chain of efmSHMT and that the biphenyl moiety and hydroxy group of (+)-SHIN-1 are important stabilisers of the interaction between efmSHMT and (+)-SHIN-1 because these moieties interact and reduce the flexibility of the 115-and 343-loops. Our findings should facilitate the development of SHMT inhibitors as therapeutic agents for bacterial, viral and parasite infections and for treating cancer. Cell cytotoxicity assays. Calu-3 and Caski cells (100 μL, 1 × 10 5 cells/mL) were cultured in RPMI (Sigma-Aldrich, St Louis, MO) containing 10% foetal bovine serum (FBS; Thermo Fisher Scientific, MA, USA), 100 units/mL penicillin G and 50 μg/mL streptomycin at 37°C in 5% CO 2 . Hep2 cells (100 μL, 1 × 10 5 cells/mL) were cultured in DMEM (Sigma-Aldrich) containing 10% FBS, 100 units/mL penicillin G and 50 μg/mL streptomycin (DMEM-FBS) at 37°C in 5% CO 2 . Each inhibitor was added to the 96-well microtiter plates and serially diluted ten-fold to produce concentrations ranging from 0.01-100 μM using the FBS-containing medium added directly to the plates. After serial dilution, 100 μL of FBS-containing medium including the cells was added to each well of the plate. The 50% cytotoxicity concentration (CC 50 ) of each compound was assessed by determining the cell viability using Cell Counting Kit-8 (DOJINDO, Kumamoto, Japan) after the cells were incubated with the serially diluted compounds for 5 days. All CC 50 determining assays were conducted in duplicate or triplicate 39,40 .

Materials and methods
Antibacterial assay with an excess concentration of biomolecules. The EC 50 values of the test compounds against E. faecium were determined in the presence of different concentrations of Ser, Gly, inosine, dT and FA using the same procedure described for the test compounds. The ratio of the control EC 50 (no competitors added) to the experimental EC 50 (competitors added) was calculated for each test substance. Plasmid construction. E. faecium was diluted to a turbidity equivalent that matched 1.0 McF standard in distilled water. The diluted sample was heated at 98°C for 10 min and then incubated at 4°C for 5 min. The sample was centrifuged, and the supernatant was collected and used as a DNA template for PCR. efmSHMT DNA was PCR-amplified using the aforementioned DNA template and an appropriate primer set (efmSHMT-F1: 5′-TATATACCCGGGGTGGTAGATTAC AAAACGTTTGAC-3′ and efmSHMT-R1: 5′-TAATAGGATCCACACCTCATAT ACTAGAGAGCATCAC-3′). The amplified DNA fragment was purified using a PCR purification kit (QIAGEN, Venlo, the Netherlands) in accordance with the manufacturer's protocols. The DNA fragment and pET-47b (Merck, Darmstadt, Germany) were digested with XmaI (NEB, MA, USA) and BamHI (TOYOBO, Shiga, Japan) at 37°C for 2 h. The digested DNAs were purified using the same PCR purification kit and ligated using T4 DNA ligase (Promega) (room temperature, 2 h). The ligated sample was transformed into JM109 cells (Wako, Osaka, Japan). Transformed cells were spread onto kanamycin-containing lysogeny broth (LB) plates and incubated at 37°C overnight. Fourteen hours later, the bacterial colonies were selected and incubated in LB, including kanamycin at 37°C overnight. The incubated samples were harvested by centrifugation, and the SHMT expression vector extracted from the incubated bacteria using a miniprep kit (QIAGEN) was sequenced (Genetic Analyzer 3500).
Protein expression and purification. The efmSHMT and ecSHMT constructs with N-terminal histidine tags and PreScission protease cleavage sites were expressed in E. coli BL21(DE3) cells. The proteins were chromatographically purified on a HisTrap column (Cytiva, Tokyo, Japan) and digested with PreScission protease (Cytiva). The solutions were then subjected to HiTrap Q (Cytiva) and Superdex 75 gel filtration chromatography steps (Cytiva). Each protein preparation was concentrated in 20 mM HEPES buffer (pH 7.5) containing 50 mM NaCl (final concentration, 15 mg/ml) and stored at 4°C.
Competitive binding assay. Protein samples were prepared as 100 μL aliquots containing 2.5 or 5 μM efmSHMT, 10 mM Gly, 100 μM Ser and each concentration (1, 2.5, 5, 10, 20, 50, 100, 200, 500, 1000, 2000, 4000 and 8000 μM) of Me-THF with 20 mM HEPES (pH 7.5) and 50 mM NaCl. After adding Me-THF, the efmSHMT solution was incubated for 30 min at room temperature. The optical density at 500 and 600 nm (OD 500 and OD 600 ) were then measured using a nanodrop spectrophotometer 24,25 . OD 600 was used as the background. After detection, each concentration of inhibitor was added to the efmSHMT solution and incubated for 30 min at room temperature. OD 500 and OD 600 were measured. The relative OD 500 intensity was calculated as relative OD 500 = (OD 500 -OD 600 )/( MAX OD 500 -OD 600 ), where MAX OD 500 is the average OD 500 value in the presence of 1000, 2000 and 4000 μM Me-THF without inhibitors. EC 50 was defined as the concentration of Me-THF that gives 50% of efmSHMT bound to Me-THF, relative OD 500 intensity of 0.5. The binding inhibition constant (K i ) was estimated using the Cheng-Prusoff equation: Dynamic light scattering measurements. Dynamic light scattering (DLS) was performed at 25°C using Zetasizer NanoS instrumentation (Malvern Instruments, Worcestershire, UK). Sample concentrations were adjusted to 0.5 mg/mL. Measurements were conducted in triplicate. Data were analysed using the algorithms included in the Zetasizer Nano software.
X-ray diffraction data collection and processing details. Diffraction data of the efmSHMT/(+)-SHIN-1 complex were collected at SPring-8 BL32XU 41 , and diffraction data of efmSHMT/Ser-(+)-SHIN-1 and efmSHMT/Gly-Me-THF complexes were collected at SPring-8 BL41XU 42 . All data were collected under cryoconditions (100 K) using the automatic collection system ZOO 43 and processed by the automatic data processing software KAMO 44 and the CCP4 43-46 software suite. The structure of the efmSHMT/(+)-SHIN-1 complex was solved using the molecular replacement method with Phaser 47 . The coordinates of the Streptococcus thermophilus SHMT (Protein Data Bank (PDB) ID: 4WXB) 48 were used as the search model. The initial phase of the efmSHMT/Ser-(+)-SHIN-1 complex was obtained from the structure of the efmSHMT/(+)-SHIN-1 complex, whereas the structure of the efmSHMT/Gly/Me-THF complex was solved by the molecular replacement method with Phaser 47 using the efmSHMT/(+)-SHIN-1 complex structure as a search model because large changes in cell parameters were observed. The model was corrected and further refined using Phenix 49,50 and Coot 51 . Data collection and refinement statistics are summarised in Table 3. Structural models in figures were generated using PyMol (http://www.pymol.org).
Relative quantitation of mRNA for expression of folate-binding proteins. E. faecium was incubated at 37°C overnight in 20 mL of MH broth with and without 2 µM of (+)-SHIN-1. Bacterial RNA was extracted using TRIzol™ Max™ Bacterial RNA Isolation Kit (Thermo Fisher Scientific, Tokyo, Japan) following the manufacturer's protocols. For the quantitative polymerase chain reaction (qPCR), the One-Step TB Green® PrimeScript PLUS RT-PCR Kit (TaKaRa, Kusatsu, Japan) was used and PCR was performed using the Thermal Cycler Dice Real Time System Lite (TaKaRa). The mRNA of D-alanine-D-alanine ligase (ddl) was used as a housekeeping RNA 52 . The primer set for qPCR is described in Supplementary  Table S1. The PCR protocol used was: reverse transcription step at 40°C for 5 min, an initial heating step at 95°C for 10 s; 40 cycles of the amplification step consisting of a 95°C denaturation period for 5 s, a 61°C annealing (58°C for TS, 60°C for SHMT and 54°C for ddl) and extension period for 30 s (1 min for ddl); and a melting phase from 65 to 95°C, following the standard protocols of the TaKaRa Dice system. The cycle threshold (C t ) was calculated by the second derivative maximum method. The relative RNA amounts of the parental strain and mutants were evaluated using the ΔΔC t method and ddl as the housekeeping RNA. Thus, the relative transcription level (RTL) was calculated using: RTL = 2 (-ΔΔCt) , ΔΔC t = ΔC t SHIN−1 -ΔC t No drug , where ΔC t represents the C t value of the RNA of each protein standardised by the C t value of the mRNA of ddl (C t ddl ). Thus, ΔC t was calculated as: ΔC t = C t -C t ddl .
Statistics and reproducibility. Statistical analysis was performed using Microsoft Excel. The p-values were calculated by t test. Average values and standard errors in each experiment were calculated from at least three independent experiments.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.