Biosynthesis of the redox cofactor mycofactocin comprises oligoglycosylation by MftF in Mycolicibacterium smegmatis

Mycofactocin (MFT) is a redox cofactor involved in alcohol metabolism of mycobacteria including Mycobacterium tuberculosis. In recent years, a preliminary biosynthetic model of MFT has been established by in-vitro studies, while the final structure of MFT remained elusive. Here, we report the discovery of MFT by metabolomics and establish a model of its biosynthesis in Mycolicibacterium smegmatis. Structure elucidation revealed that MFT is decorated with up to nine β-1,4-linked glucose residues. Dissection of biosynthetic genes demonstrated that the oligoglycosylation is catalyzed by the glycosyltransferase MftF. Furthermore, we confirm the cofactor function of MFT by activity-based metabolic profiling using the carveol dehydrogenase LimC and show that the MFT pool expands during cultivation on ethanol. Our results close an important gap of knowledge, will guide future studies into the physiological roles of MFT in bacteria and may inspire its utilization as a biomarker or potential drug target to combat mycobacterial diseases.


Introduction
Coenzymes are small molecules that are indispensable for the catalytic activity of many enzymes.
While coenzymes like NAD + or FAD are ubiquitous in nature and are essential for the core metabolism of all forms of life, specialized cofactors like pyrroloquinoline quinone (PQQ) 1 and coenzyme F 420 2 are restricted to certain microbial phyla, but typically involved in extraordinary metabolic processes like methylotrophy, methanogenesis, or detoxification processes. Mycobacteria are particularly rich in unusual redox cofactors and antioxidants that contribute to redox balance and metabolic plasticity. For instance, mycothiol 3,4 or ergothioneine 5 protect Mycobacterium tuberculosis from oxidative stress and support detoxification pathways while coenzyme F 420 is involved in a response to nitrosative stress 6 . Pathogenic mycobacteria are exposed to a variety of carbon sources, reactive oxygen species, and reactive nitrogen species inside the host organism. Furthermore, persistent M. tuberculosis may enter a dormant state characterized by low metabolic activity and reduced susceptibility to antibiotics, where survival or death are governed by complex redox signaling pathways and alterations of electron transport processes.
Moreover, some antimycobacterial drugs are administered as prodrugs and will only develop bioactivity upon transfomation, e.g. protomanid 7 , which is activated by a coenzyme F 420 -dependent reductase 8 .
Hence, unusual cofactors that assist such biochemical processes are critical for survival, virulence, and drug resistance of M. tuberculosis 9 . The emergence of extensively drug-resistant tuberculosis (XDR-TB) is one of the major threats to human health worldwide and highlights the urgency of the exploration of metabolic systems that may influence the fitness of a pathogen during infection and modulate antibiotics activity.
Mycofactocin (MFT) 10 is a putative redox-cofactor whose existence has been postulated on the basis of comparative genomics and bioinformatics 11 . Its final molecular identity and structure, however, have remained elusive to date. The MFT biosynthetic gene cluster is highly conserved and wide-spread among mycobacteria. The inactivation of the MFT gene locus in the model species Mycolicibacterium smegmatis (synonym: Mycobacterium smegmatis) as well as M. tuberculosis resulted in the inability of the mutants to utilize ethanol as a sole source of carbon and further disturbances of mycobacterial redox homeostasis were revealed 12 . Involvement of MFT in methanol metabolism was reported as well 13 . These recent results strongly support the hypothesis that MFT is a redox cofactor and might represent a fitness factor of mycobacteria during some stages of infection.
The architecture of the MFT gene cluster (Fig. 1a) suggested that the resulting natural product is a ribosomally synthesized and post-translationally modified peptide (RiPP) 14 . Several in-vitro studies have contributed to a biosynthetic model of MFT (Fig. 1b): The precursor peptide MftA consisting of 32 amino acids is produced by the ribosome and bound by its chaperone MftB. Subsequently, the terminal core peptide consisting of Val and Tyr is oxidatively decarboxylated and cyclized by the radical SAM enzyme MftC 15,16,17 . The resulting cyclic core structure is released by the peptidase MftE 18 forming 3-amino-5-[(p-hydroxyphenyl)methyl]-4,4-dimethyl-2-pyrrolidinone (AHDP) 19 . Just recently, it was shown that MftD, an enzyme homologous to the L-lactate dehydrogenase LldD2 20 , catalyzes the oxidative deamination of AHDP to yield pre-mycofactocin (PMFT) 21 . The same study demonstrated by voltammetry that the a-keto amide moiety of PMFT is redox-active and can be reduced to PMFTH 2 (midpoint potential: -255 mV). Efficient reduction was also achieved by the action of carveol dehydrogenase using carveol as an electron donor in vitro 21 . Therefore, PMFT likely represents the redoxcenter of MFT, as riboflavin is the redox-active core of FMN and FAD.
Although these current hypotheses are plausible, all of these known metabolic intermediates have just been observed in vitro and could therefore represent artifacts. The verification of their relevance in vivo was urgently desired and the final steps of MFT biosynthesis as well as the function of the mftF gene awaited experimental elucidation. In this study, we confirm the current biosynthetic model of MFT in vivo and detected several novel oligoglycosylated MFT congeners. We also show that the sugar chain is a b-1,4-glucan and provide first evidence that glycosylation is performed by the glycosyltransferase MftF.
Finally, we show dependence of MFT formation on ethanol and corroborate its cofactor function by activity-based metabolic profiling.

Discovery of mycofactocins by metabolomics
In order to identify potential mycofactocin congeners in mycobacteria, we used the fast-growing and weakly pathogenic species M. smegmatis MC 2 155 as a model organism and developed a metabolomics approach combining metabolic induction and labeling to specifically trace MFT congeners. Assuming that MFT production would be stimulated by alcohols, we cultivated bacteria in media containing 10 g L -1 ethanol. Furthermore, we used stable isotope labeling to obtain candidate molecules compatible with the proposed biosynthetic pathway: Since the C-terminal core peptide of MftA is composed of Val and Tyr, we reasoned that MFT congeners could be specifically labeled by feeding L-Val-13 C 5 and L-Tyr-13 C 9 .
Intracellular contents were extracted and analyzed by liquid chromatography coupled with high-resolution mass spectrometry (LC-MS). Compounds were detected by in-silico grouping of co-eluting isotopic peaks and adducts (feature finding). Afterwards, 13 C-labeled compounds were deduced computationally (Supplementary Table 1). According to the established biosynthetic pathway we expected 13 carbons to remain 13 C-labeled after oxidative decarboxylation of the Val-Tyr core peptide. We therefore searched for compounds that displayed an exchange of exactly 13 carbons, resulting in a mass shift of +13.04362 Da ( Supplementary Fig. 1). After removal of low-quality hits, this approach revealed a list of only twelve candidate compounds. Strikingly, the exact mass and proposed sum formula of three of these compounds corresponded to known intermediates of MFT, namely AHDP, PMFT as well as PMFTH 2 , indicating that our strategy was viable. In addition to these compounds, several labeled molecules with increasing molecular weight were detected. Some co-eluting candidates with a mass difference of +17.02654 could be explained as NH 4 + adducts of each other. The remaining nine candidate compounds (Table 1) were grouped based on their chromatographic retention times, eluting closely to either PMFT or PMFTH 2 (approx. 7.2 min and 6.9 min, respectively). Intriguingly, members of the two groups could be arranged in pairs with a mass difference of two hydrogen atoms leading us to assume that each group represented derivatives of either PMFT or PMFTH 2 . Thus, we termed these two clusters of molecules mycofactocinones (MFT) and mycofactocinols (MFTH 2 ), respectively. Notably, some mycofactocinols eluted as two chromatographically separated isomers. For instance, the dominant PMFTH 2 peak eluted at 6.8 min, while the minor isomer eluted at 6.5 min. These two compounds displayed highly similar MS/MS spectra ( Supplementary Fig. 2) and most likely represent tautomeric forms. For the sake of simplicity, only the more prevalent isomer was considered during metabolomics studies.
During further experiments, we performed MS/MS networking (Fig. 2), an approach that clusters compounds based on similarity of their MS/MS fragmentation pattern and therefore potentially related chemical scaffolds 22 . Interestingly, candidates retrieved from 13  MMFT-2H 2 (574.24640) suggesting that the methyl group is present in the second sugar moiety (Fig. 2a).

Structure elucidation of the oligosaccharide moiety
To determine the exact structure of elongated mycofactocins we conducted large-scale fermentations (50 L) in a fermentor and harvested the resulting biomass. Cell lysis and MS-guided purification using a solid-phase extraction procedure resulted in an enriched fraction of the dominant mycofactocin. This species exhibited the same exact mass and fragmentation pattern as MMFT-2H 2 , but eluted at a slightly shifted retention time ( Supplementary Fig. 3). We therefore named this species MMFT-2bH 2 .
Due to the low yields of individual MMFT derivatives, co-elution of contaminants and intrinsic purification problems associated with oligoglycosylated compounds, structural analysis by nuclear magnetic resonance (NMR) was not possible at this stage. Thus, we used enzymatic degradation to obtain structural information about the oligoglycoside chain. Gratifyingly, cellulase (β-1,4-glucanase) degraded the sugar chain of mycofactocin species (n>2), while amylase (a-1,4-glucanase) did not exhibit any effect ( Fig. 3a). This finding strongly suggested that the oligosaccharide chain represents a β-1,4-glucan tuberculosis as a constituent of biofilms after exposure to reductive stress. 24 The production of methylated glucans, like 6-O-methylglucose lipopolysaccharides (MGPL), albeit with a-1,4 linkage, is well described in Mycobacteria 25 , 2-O-methylglucose appears to be less common. To the best of our knowledge, MFT and mycothiol 4 are the only cofactors that are decorated with sugar moieties.

Glycine-derived intermediates of MFT biosynthesis
Surprisingly, the MS/MS network ( Fig. 2a) revealed two additional compounds (m/z 1426.54 and 1588.59)) with an unusual mass shift compared to the MFT-n(H 2 ) candidates. Their mass differences and MS/MS spectra indicated that they represented hepta-and octaglycosylated species sharing a head moiety closely related to PMFT and PMFT(H 2 ). The molecular masses and MS/MS spectra of the compounds could be explained by the assumption that the aglycon corresponded to glycyl-AHDP (GAHDP) and these compounds represented the oligoglycosylated forms GAHDP-7 and GAHDP-8. Since the VY core peptide of MftA is preceded by a glycine residue at its N-terminal side it appeared highly likely that the GAHDP-n species corresponded to premature cleavage products of the MftC-processed precursor peptide.
To corroborate this hypothesis, we fed M. smegmatis cultures with a combination of fully 13 C-labeled Gly-13 C 2 , Val-13 C 5 , and Tyr- 13 Table 2).

Dissection of MFT biosynthesis
In order to test if all of the MFT candidate compounds were related to MFT biosynthesis, we started to investigate mutants (DmftC, DmftD, DmftE, DmftF) of the MFT biosynthesis pathway for the production of candidate molecules. Since MFT mutants, except for DmftE, do not grow on ethanol as a sole source of carbon 12 , we first cultivated bacteria on the surface of cellulose filters on agar plates containing 10 g L -1 glucose as a carbon source. The filters were then transferred to treatment plates containing 10 g L -1 ethanol and incubated overnight. This strategy allowed for induction of MFT production in the wild type  The DmftE mutant was able to produce (M)MFT-nH 2 candidates, albeit in significantly lower amounts, explaining the previously unexpected phenotypic observation that the DmftE mutant was able to grow on ethanol, but slower than WT 12 . Intriguingly, the pool of GAHDP-n was strongly increased in the DmftE strain (Fig. 4b). We thus conclude that MftE can be complemented by an unknown peptidase present in the metabolic background of mycobacteria. Theoretically, an aminopeptidase would be sufficient to degrade the N-terminus of MftA, releasing the AHDP-like core. Peptidases encoded outside the biosynthetic gene cluster have been observed in other RiPP biosyntheses as well 26 . However, the removal of the glycine residue is an apparent bottle-neck of the alternative maturation pathway in

M. smegmatis.
In full agreement with the in-vitro finding that MftD consumes AHDP to form PMFTH 2 21 , all metabolites downstream of (M)AHDP-n were abrogated in the DmftD strain, whereas AHDP-n and GAHDP-n accumulated (Fig. 4). The fact that GAHDP-n increased might suggest that the MftE step is impeded in the absence of MftD as well. Genetic dysregulation or cooperative effects between the two enzymes, like complex formation and substrate channeling, might account for this result.

Glycosylation of MFT is mediated by MftF
It has been speculated that the putative glycosyltransferase MftF catalyzes a final glycosylation of PMFT to yield the mature cofactor 21 . Our results at this point showed that multiple glucose residues are attached to the aglycon in vivo. Furthermore, glycosylation appeared already at an early stage as mirrored by the presence of the glycosylated (G)AHDP-n series. In order to link oligoglycosylation to a given gene product, we analyzed the DmftF mutant for the production of glycosylated MFT congeners. Indeed, all glycosylated MFT congeners were abolished in the DmftF metabolome. Unexpectedly, DmftF mutants additionally ceased to produce the aglycons PMFT and PMFTH 2 . MftF did, however, produce trace amounts of AHDP, thus showing that at least residual MftC activity was present in the mutant (Fig. 4b).
To exclude polar effects, we complemented DmftF by re-introduction of the mftF gene under control of the mftA promotor. The restoration of the full MFT metabolite spectrum (Supplementary Table 2 Sequence alignment ( Supplementary Fig. 21a) showed a high degree of sequence conservation among mycobacterial species and other actinomycetes (e.g., 92% similarity to MftF of M. tuberculosis H37Rv).
Prediction of transmembrane domains revealed a single helix spanning residues 324 -346 with the Nterminus being located outside of the membrane Supplementary Fig. 21b). The MMFT biosynthetic machinery, however, appears not to be fully encompassed within the MFT cluster since no methyltransferase was found. Future studies are warranted to identify the enzymes involved in MFT oligosaccharide methylation.

Cofactor role of mycofactocin
After discovery of the full set of mycofactocins, we examined to which extent their production was actually dependent on the presence of ethanol in culture media. We therefore systematically compared the metabolome of M. smegmatis WT after ethanol treatment with glucose controls. (Supplementary Table 3).
The results demonstrated that all MFT congeners and intermediates were strongly upregulated upon cultivation on ethanol (median: 34-fold upregulation). (Fig. 5a). These data perfectly support a recent report that MFT is involved in alcohol metabolism 12 .
Finally, we sought to confirm that the MFT congeners identified in this study are actually coenzymes of MFT-dependent enzymes. To assess this question, we turned to activity-based metabolic profiling 27 .
We incubated the extracted metabolome of M. smegmatis with the recombinant carveol dehydrogenase LimC from Rhodococcus erythropolis ( Supplementary Fig. 22), a nicotinoprotein with a non-exchangable NADH cofactor 28 . This enzyme was proposed to require MFT as an external electron acceptor 11 . A recent study showed that carveol dehydrogenase from M. smegmatis was able to reduce PMFT to PMFTH 2 using carveol and internally bound NADH as an electron donor 21 . Likewise, we observed full reduction of all mycofactocinones to mycofactocinols (Fig. 5b, Supplementary Table 4) by LimC when combined with carveol as a substrate. Controls lacking enzyme or substrate showed weak and no turnover, respectively.
The low turnover by LimC alone can be explained by internally bound NADH as reported before 21 . Both the aglycon PMFT as well as the oligoglycosylated MFT-n and MMFT-n species were completely turned over, while redox-inactive AHDP congeners remained unaffected. These data further validate the notion that all MFT candidates presented here are bona-fide mycofactocins with full cofactor function.

Conclusion
The proposed redox cofactor mycofactocin has attracted considerable interest since it was postulated by bioinformatics. Despite recent progress made by in-vitro studies, evidence for mycofactocin congeners in living microorganisms has been missing so far. Our metabolomics approach combined with stable isotope labeling, induction by ethanol, as well as genetic dissection of the biosynthetic pathway turned out to be a powerful approach to identify MFT congeners in mycobacteria. We discovered that MFT is decorated with oligosaccharides consisting of up to nine b-1,4-linked glucose units. Analyses of DmftF mutants and complement strains revealed that MftF is the glycosyltransferase responsible for the oligoglycosylation observed. Mycofactocins can be isolated in oxidized (mycofactocinones) and reduced forms (mycofactocinols) and are co-substrates of enzymatic reduction by carveol dehydrogenase. These data provide strong evidence that mycofactocins are indeed redox cofactors as proposed earlier 11,12,21 . We therefore conclude that we have finally discovered the family of compounds that was tentatively called "mycofactocin" and thus close an important gap of knowledge in the field. Our results will guide further studies into the occurrence, physiological role, and biochemistry of mycofactocins in microorganisms.
Finally, we are confident that this work will inspire future efforts to exploit mycofactocin as a disease marker or as a potential drug target for the treatment of tuberculosis and other mycobacterial infections.

Isotopic labeling of M. smegmatis
A saturated pre-culture of M. smegmatis MC 2 155 WT in LB broth was used to inoculate 25 mL of HdB medium supplemented with 0.5 g L -1 tyloxapol and 10 g L -1 ethanol to an initial optical density at 600 nm (OD 600 ) of 0.1. Cultivations contained 1 mM L-tyrosine-13 C 9 (99% atom purity, Cortecnet) and 1 mM L-valine-13 C 5 (99% atom purity, Merck) and were conducted in four replicates at 37 °C and 180 rpm for 24 h. Cultures (10 mL of a 10-fold dilution) were poured onto sterile regenerated cellulose filters (0.2 µm, Sartorius) previously conditioned with water. The biomass was repeatedly washed with sterile water and transferred to HdB agar plates with 10 g L -1 ethanol and with either light or heavy Lvaline and L-tyrosine supplementation, as appropriate. Filters inoculated with sterile media were used as control. Filters were incubated at 37 °C for 48 h. Directly after the incubation period, the filters were extracted and subjected to LC-MS measurements as described below. Data analysis was performed with the Stable Isotope Labeling workflow of Compound Discoverer 3.0 (Thermo Scientific) allowing for a maximum exchange of 16 13 C atoms. Independent analyses were performed for the lower and higher scan ranges. A minimal peak intensity cut-off of 10 4 was defined for compound detection. Compounds with the same mass (± 5 ppm) and eluting within 0.2 min from each other were grouped. Molecules were considered candidates potentially comprising the decarboxylated Val-Tyr core peptide (i.e. AHDP moiety) if a relative 13 C-exchange rate higher than 50% was observed. Low abundance compounds (area <1000) and candidates containing a high proportion of contaminating masses were disregarded.

Comparative metabolomics studies
Cultures of M. smegmatis MC 2 155 WT as well as DmftC, DmftD, DmftE, DmftF, DmftC-Comp and DmftF-Comp growing in LB supplemented with 0.05% of Tween80 were used to inoculate sterile regenerated cellulose filters as described before, standardizing all the cultures to the same concentration.
The filters were incubated in HdB supplemented with 10 g L -1 glucose at 37 °C for 18h. Afterwards, the filters were transferred to a new HdB plate supplemented either with 10 g L -1 of glucose or 10 g L -1 ethanol and incubated at 37°C for 18h. This study was carried out in triplicates and filters incubated with media were used as control (blank). Directly after the incubation period, the filters were further extracted and subjected to LC-MS measurements as described below. Targeted studies were performed using the Expected Compounds node in Compound Discoverer 3 with an Expected Compounds table including AHDP, PMFT, PMTH 2 , MFT-1 and MFT-1H 2 allowing for multiple glycosylation and methylation events. Compounds from different runs with the same mass (< 5 ppm deviation) and eluting within 0.2 min from each other were grouped. Median of areas under the curve of three replicates was used to compute ratios between groups.

Metabolite extraction and LC-MS measurements
Filters were recovered with sterile tweezers and placed on 20 mL chilled extraction mixture

Bioinformatics analysis of the MftF primary structure
The primary protein structure of several MftF homologues was downloaded from the NCBI database.
Sequences were aligned using the MUSCLE algorithm 33 implemented in Geneious Prime (2019.1.1).
Prediction of transmembrane domains was performed using the TMHMM 2.0 webserver 34 . Classification of MftF was performed by the carbohydrate-active enzymes database (CAZy) 35 .

Heterologous production and purification of LimC
The limC gene encoding the carveol dehydrogenase LimC 28 from Rhodococcus erythropolis DCL14 was obtained as a codon optimized synthetic construct inserted in vector pET28, in-frame with the N-terminal hexahistine tag (plasmid pLAPO4, see Supplementary Fig. 23a). A single colony of E. coli BL21(DE3) freshly transformed with plasmid pLAPO4 was inoculated in 5 mL LB with kanamycin (50 µg mL -1 ) and cultured overnight at 37 °C and 210 rpm. A main culture (100 mL) was inoculated in LB media amended with the same antibiotic and cultured until an OD 600 of 0.5 -0.6. At this point expression of limC was induced with IPTG (0.5 mM) and the temperature decreased to 16

Characterization of sugar composition
See Supplementary Information, Section 1.

Gas chromatography -mass spectrometry (GC-MS)
GC-MS analysis was conducted using an Agilent 6890 Series gas chromatograph coupled to an Agilent