Functional Screening of Metagenome and Genome Libraries for Detection of Novel Flavonoid-Modifying Enzymes

Abstract

The functional detection of novel enzymes other than hydrolases from metagenomes is limited since only a very few reliable screening procedures are available that allow the rapid screening of large clone libraries. For the discovery of flavonoid-modifying enzymes in genome and metagenome clone libraries, we have developed a new screening system based on high-performance thin-layer chromatography (HPTLC). This metagenome extract thin-layer chromatography analysis (META) allows the rapid detection of glycosyltransferase (GT) and also other flavonoid-modifying activities. The developed screening method is highly sensitive, and an amount of 4 ng of modified flavonoid molecules can be detected. This novel technology was validated against a control library of 1,920 fosmid clones generated from a single Bacillus cereus isolate and then used to analyze more than 38,000 clones derived from two different metagenomic preparations. Thereby we identified two novel UDP glycosyltransferase (UGT) genes. The metagenome-derived gtfC gene encoded a 52-kDa protein, and the deduced amino acid sequence was weakly similar to sequences of putative UGTs from Fibrisoma and Dyadobacter. GtfC mediated the transfer of different hexose moieties and exhibited high activities on flavones, flavonols, flavanones, and stilbenes and also accepted isoflavones and chalcones. From the control library we identified a novel macroside glycosyltransferase (MGT) with a calculated molecular mass of 46 kDa. The deduced amino acid sequence was highly similar to sequences of MGTs from Bacillus thuringiensis. Recombinant MgtB transferred the sugar residue from UDP-glucose effectively to flavones, flavonols, isoflavones, and flavanones. Moreover, MgtB exhibited high activity on larger flavonoid molecules such as tiliroside.

INTRODUCTION

For more than a decade, metagenome research has demonstrated that it is a powerful tool for the discovery of novel biocatalysts and other valuable biomolecules by using either function- or sequence-based screening technologies (1–3). Sequence-based approaches allow the identification of candidate genes. In particular, the development of next-generation sequencing (NGS) technology and improved bioinformatic tools have significantly advanced this methodology (4). However, a major drawback of sequence-based screening technologies is that they do not allow direct conclusions about the functionality and biochemical parameters of the encoded enzymes. Furthermore, sequence-based searches are limited to the identification of homologs of already known motifs (5). Yet another problem associated with the sequence-based approach is that it often reveals only partial genes, which make subsequent expression and detailed biochemical analysis of the gene products difficult if not impossible. In contrast, the function-driven approach is usually much slower and more labor-intensive and costly but results in the detection of complete and active enzyme clones. It is of course well known that function-driven metagenomics is hampered due to the problems of expressing genes and incorrect processing of enzymes (6–8). However, the function-based approach allows the identification of truly novel enzymes, and it allows a first judgment on the actual enzyme activities and physicochemical parameters even during the screening process.

It is not surprising that the majority of metagenome-derived enzymes that have been characterized biochemically mainly originated from function-based screenings (9, 10). Interestingly, the majority of biocatalysts that have been identified through functional approaches are hydrolytic enzymes, mainly esterases and glycoside hydrolases (11, 12). This is perhaps linked to the simple plate-based screening procedures required for rapid detection of these hydrolytic enzymes (13). In this context it is noteworthy that function-based screening technology is limited by the availability of sensitive and reliable assays for enzymes other than hydrolases that are of importance for biotechnology. Since the overall hit rates are usually low, metagenome screens often require high-throughput screening (HTS) technologies to be efficient, and the screenings need to be done under nearly production conditions (6). Further, enzyme screens often require complex substrates and sophisticated chromogenic assays as well as high-performance liquid chromatography (HPLC) or similar analytical methods. Clearly, the setup and development of novel function-driven metagenome screening assays are very tedious and time-consuming. This may be one reason why only a few function-based metagenome screening techniques have been developed during the last decade that focus on enzymes other than hydrolases and on those with relevance to biotechnological processes (14–20). Thus, there is an urgent need to develop function-based screening methods for genes and enzymes that belong to enzyme classes other than hydrolases and that are of relevance to biotechnology.

Flavonoids, as natural substances in fruits and vegetables, are part of our daily nutrition. They are well known for their antioxidative and radical scavenging nature and even more for having various beneficial effects on human health (21). Because of these broad effects, there is an increasing demand for specific flavonoids in the cosmetic industry and the pharma- and nutraceutical industries (22–24). A major problem in meeting this demand arises from their limited availability. Flavonoids are exclusively produced in plants at low levels. The extraction is linked to the use of large quantities of solvents, and the chemical modification is not easily accomplished due to their rather complex structures (25).

The regio-specific modification of flavonoids remains difficult as the directed chemical modification mostly fails. Thus, flavonoid-modifying enzymes have gained interest as they can mediate the regio- and stereochemical modification of flavonoids (26). In particular, the specific glycosylation of flavonoids is the focus of research to influence water solubility and bioavailability of the polyphenolics (27, 28). Enzymes that catalyze this reaction are glycosyltransferases (GTs). Generally, GTs mediate the transfer of sugar residues from a donor substrate to acceptor molecules. Based on their sequence similarities, GTs are currently classified into 94 families (29). GT family 1 (GT1) comprises enzymes that catalyze the glycosylation of small lipophilic molecules (30). These enzymes (EC 2.4.1.x) that use a nucleotide-activated donor belong to the UDP-glycosyltransferase (UGT) superfamily and are also referred to as Leloir enzymes (31, 32). Glycosyltransferases acting on flavonoids also belong to GT1 (33). Enzymes of GT1 possess a GT-B fold structure and present an inverting reaction mechanism concerning the linkage of the transferred sugar moiety (34). Until now very few flavonoid-acting GT1s of prokaryotic origin have been identified and characterized in detail. The currently known flavonoid-accepting UGTs derived from Gram-positive bacteria all belong to the macroside glycosyltransferase (MGT) subfamily and originate from bacilli and streptomycetes (35–37). Furthermore, a single flavonoid-acting UGT derived from the Gram-negative Xanthomonas campestris is known (38).

In the current publication we report on semiautomated thin-layer chromatography (TLC) screening of clone pools from metagenome libraries. The novel method allows the rapid identification of flavonoid-modifying enzyme clones. Using this technology, we have screened more than 40,000 fosmid clones and thereby identified two positive clones that showed significant flavonoid GT activities. The two novel enzymes, designated MgtB and GtfC, belong to GT family 1 and are highly active on flavonoids and similar molecules. While MgtB is highly similar to a hypothetical Bacillus thuringiensis MGT, GtfC is weakly similar to a hypothetical protein from Fibrisoma limi.

MATERIALS AND METHODS

Bacterial strains, plasmids, and chemical reagents.

Bacterial strains and plasmids used in the present work are listed in Table S1 in the supplemental material, and primers are listed in Table S2. If not otherwise stated, Escherichia coli was grown at 37°C in LB medium (1% tryptone, 0.5% yeast extract, 0.5% NaCl) supplemented with appropriate antibiotics. Bacillus isolates were grown at 30°C in the same medium. All chemical reagents used were of analytical-laboratory grade. Polyphenolic substances were purchased from the following companies located in Germany: Merck KGaA, Darmstadt; Carl Roth GmbH, Karlsruhe; Sigma-Aldrich, Heidelberg; and Applichem GmbH, Darmstadt. Additional flavonoids were ordered from Extrasynthese (Lyon, France). Stock solutions of the polyphenols were prepared in dimethyl sulfoxide (DMSO) in concentrations of 100 mM.

Isolation of DNA and fosmid library construction.

Bacillus sp. strain HH1500 was originally isolated from a soil sample of the botanical garden of the University of Hamburg. DNA from Bacillus sp. HH1500 was isolated using a peqGOLD Bacterial DNA Kit (PEQLAB Biotechnologie GmbH, Erlangen, Germany) by following the manufacturer's protocol. The sample for the construction of the elephant feces library was derived from the Hagenbeck Zoo (Hamburg, Germany). Fresh feces of a healthy 6-year-old female Asian elephant (Elephas maximus) named Kandy were taken and stored at −20°C in TE buffer (10 mM Tris-HCl, 1 mM EDTA, pH 8) containing 30% (vol/vol) glycerol until DNA extraction. For DNA extraction a QIAamp DNA Stool Mini Kit (Qiagen, Hilden, Germany) was used. The kit was applied according to the manufacturer's protocol. As recommended, we increased the incubation temperature in ASL buffer (Qiagen) to 95°C. Isolation of DNA from Elbe river sediment was performed with sediment samples from the tidal flat zone of the river Elbe near Glückstadt (Germany) at low tide (53°44′40″N, 9°26′14″E). Environmental DNA was extracted using an SDS-based DNA extraction method published by Zhou and coworkers (39).

Construction of the genomic and metagenomic libraries in E. coli EPI300 cells harboring fosmid pCC1FOS was achieved with a CopyControl Fosmid Library Production Kit (Epicentre Biotechnologies, Madison, WI) according to the manufacturer's protocol using minor modifications as previously published (40). Clones were transferred into 96-well microtiter plates containing 150 μl of liquid LB medium with 12.5 μg/ml of chloramphenicol and allowed to grow overnight. Libraries were stored at −70°C after the addition of 100 μl of 86% glycerol to each microtiter well. The genomic fosmid library of Bacillus sp. HH1500 comprised 1,920 clones; a total of 35,000 clones were obtained for the river Elbe sediment library, and the elephant feces library encompassed a total of 20,000 clones. All libraries contained fosmids with average insert sizes of 35 kb.

Molecular cloning strategies.

Fragments of pCC1FOS fosmids were subcloned into pBluescript II SK+ vector using HindIII according to the restriction of the fosmid clones pFOS4B2 and pFOS144C11. The resulting plasmids were designated pSK4B2 and pSK144C11, respectively. Further subcloning of pSK144C11-derived fragments was achieved in pTZ19R-Cm with restriction enzymes EcoRI and PstI. The obtained clones were designated pTZ144E and pTZ144P, respectively. E. coli DH5α was transformed with the plasmids by heat shock, and the plasmids carrying subclones were identified by blue-white screening on LB agar plates containing 10 μM 5-bromo-4-chloro-indolyl-β-d-galactopyranoside (X-Gal) and 400 μM isopropyl-β-d-thiogalactopyranoside (IPTG) after overnight growth. Different clones were analyzed by plasmid purification, followed by enzymatic digestion and agarose gel electrophoresis and/or DNA sequencing.

PCR amplification of open reading frames (ORFs) was performed with fosmid DNA as a template. The reactions were performed in 30 cycles. To amplify mgtB, the primers mgt1-XhoI-for and mgt1-XhoI-rev were used, inserting XhoI endonuclease restriction sites 5′ and 3′ of the ORF (see Table S2 in the supplemental material). For cloning of gtfC, the primer pair gtf-Nde-for and gtf-Bam-rev was used, inserting an NdeI site, including the start codon, 5′ of the ORF and a BamHI site 3′ of the ORF (see Table S2). PCR fragments were ligated into pDrive using a Qiagen PCR Cloning Kit (Qiagen, Hilden, Germany) and cloned into E. coli DH5α. Resulting clones, designated pDmgtB and pDgtfC, respectively, were analyzed for activity in biotransformation and by DNA sequencing for the correct insert. Ligation of mgtB and gtfC into expression vector pET19b (Merkc KGaA, Darmstadt, Germany) was achieved using the inserted endonuclease restriction sites of each ORF. Plasmids containing the correct insert were designated pET19mgtB and pET19gtfC, respectively. E. coli DH5α clones harboring the desired plasmids were detected by direct colony PCR using a T7 terminator primer and mgt1-XhoI-for to confirm insertion of mgtB, and the T7 terminator primer and gtf-Nde-for were used to verify gtfC. Additionally, the inserts of pET19mgtB and pET19gtfC were sequenced using T7 promoter and T7 terminator primers (see Table S2) to verify the constructs.

Overproduction and purification of enzymes.

For overproduction of decahistidine (His₁₀)-tagged proteins, E. coli BL21(DE3) was transformed with pET19b constructs. An overnight preculture was harvested by centrifugation, and 1% was used to inoculate an expression culture. Cells carrying pET19mgtB were grown at 22°C until an optical density at 600 nm (OD₆₀₀) of 0.7. The culture was transferred to 17°C and induced by 100 μM IPTG. After 16 h, the culture was harvested by centrifugation at 7,500 × g at 4°C. Cells were resuspended in 50 mM phosphate-buffered saline (PBS) with 0.3 M NaCl at pH 7.4 and disrupted by ultrasonication with an S2 Sonotrode in a UP200S instrument (Hielscher, Teltow, Germany) at a cycle of 0.5 and an amplitude of 75%.

The overproduction of decahistidine-tagged GtfC was induced at 37°C at an OD₆₀₀ of 0.6 with 100 μM IPTG. Cells were then incubated f for 4 h, harvested, and lysed as stated above for MgtB.

Crude cell extracts were centrifuged at 15,000 × g and 4°C to sediment the cell debris. The clarified extracts were loaded on 1-ml HisTrap FF Crude columns using an ÄKTAprime Plus system (GE Healthcare). The enzymes were purified according to the manufacturer's protocol for gradient elution of His-tagged proteins. Eluted protein solutions were dialyzed twice against 1,000 volumes of 50 mM PBS, pH 7.4, with 0.3 M NaCl at 4°C. The purification was analyzed by 12% SDS-PAGE. The concentration of protein was determined by the Bradford method using Roti-Quant (Carl Roth GmbH, Karlsruhe, Germany).

Biotransformations and biocatalyses.

For the detection of flavonoid modifications in bacteria, we used a biotransformation approach. Cultures were grown in LB medium with appropriate antibiotics overnight. Expression cultures were prepared as stated above for overproduction of enzymes. The cells were sedimented by centrifugation at 4,500 × g and resuspended in 50 mM sodium phosphate buffer, pH 7, supplemented with 1% (wt/vol) α-d-glucose. Biotransformations with a final concentration of 100 μM flavonoid inoculated from stock solutions of 100 mM in DMSO (i.e., 0.1%) were incubated in Erlenmeyer flasks at 30°C and 175 rpm up to 24 h. Samples of 4 ml were withdrawn and acidified with 100 μl of 1 M H₃PO₄ aqueous (aq) for extraction in 2 ml of ethyl acetate (EtOAc). They were shaken for 1 min and phase separated by centrifugation at 2,000 × g and 4°C. The supernatant was applied in TLC analysis. For quantification, samples of 100 μl were taken and dissolved 1/10 in ethyl acetate-acetic acid (3:1). These acidified ethyl acetate samples were centrifuged at 10,000 × g. The supernatant was used for quantitative TLC analysis as stated below.

Fosmid clones were grown in 96-deep-well plates overnight. Clones were joined in 96, 48, 8, or 6 clones per pool. The pools were harvested by centrifugation at 4,500 × g and resuspended in 50 ml of LB medium containing 12.5 μg/ml chloramphenicol (CopyControl Autoinduction Solution; Epicentre, Madison, WI) (5 mM arabinose final concentration) and 100 μM flavonoid for biotransformation. As an alternative to deep-well plates, clones were precultured on agar plates. After overnight incubation, the colonies where washed off with 50 mM sodium phosphate buffer, pH 7, harvested by centrifugation, and resuspended as outlined above. The biotransformations were incubated in 300-ml Erlenmeyer flasks at 30°C with shaking at 175 rpm. Single clones were tested analogously but precultured in 5 ml of LB medium and resuspended in 20 ml of biotransformation medium in 100-ml flasks. Samples of 4 ml were taken from the reaction products after 16, 24, and 48 h, acidified with 40 μl HCl aq, and prepared for TLC analysis as stated above. Positive pools were verified in a second biotransformation and then systematically downsized to detect the corresponding hit in a smaller pool until the responsible single clone was identified.

Biocatalytic reaction mixtures of 1 ml contained 5 μg of purified His-tagged enzyme, and reactions were performed in 50 mM sodium phosphate buffer, pH 7, at 37°C. UDP-α-d-glucose or UDP-α-d-galactose was added to a final concentration of 500 μM as a donor substrate from 50 mM stock solutions in 50 mM sodium phosphate buffer, pH 7. Acceptor substrates were used in concentrations of 100 μM and were added from stock solutions of 100 mM in DMSO, leading to a final content of 0.1% in the reaction mixture. The reaction was stopped by dissolving 100 μl of reaction mixture 1/10 in ethyl acetate-acetic acid (3:1). These samples were used directly for quantitative TLC analysis.

TLC analyses.

The extracts transferred into HPLC flat-bottom vials were used for TLC analysis. Samples of 20 μl were applied on 20- by 10-cm² high-performance thin-layer chromatograph (HPTLC) silica gel 60 F₂₅₄ plates (Merck KGaA, Darmstadt, Germany) versus 200 pmol of reference flavonoids. To avoid carryover of substances, i.e., to prevent false positives, samples were spotted with double syringe rinses in between by an Automatic TLC Sampler 4 instrument (ATS 4; Camag, Muttenz, Switzerland). The sampled TLC plates were developed in ethyl acetate-acetic acid-formic acid-water (100:11:11:27) (Universal Pflanzenlaufmittel, or universal plant solvent) (41). After band separation, the TLC plates were dried in an oven at 80°C for 5 min. The absorbance of the separated bands was determined densitometrically depending on the absorbance maximum of the applied educts at 285 to 370 nm using a deuterium lamp in a TLC Scanner 3 (Camag, Muttenz, Switzerland). Subsequently, the substances on developed TLC plates were stained by either dipping or spraying the plates in a 1% (wt/vol) methanolic solution of Naturstoff reagent A, containing diphenyl boric acid β-aminoethyl ester (42), available from Carl Roth GmbH, Karlsruhe, Germany. After immediate drying with a hot air fan, the TLC plates were dipped in or sprayed with a 5% (wt/vol) solution of polyethylene glycol 4000 in ethanol (70%, vol/vol). For dipping, a chromatogram immersion device (Camag, Muttenz, Switzerland) was used. After complete drying the bands were visualized at 365 nm with a UV hand lamp and photographed. Alternatively, fluorescence of the bands was determined densitometrically by the TLC Scanner 3 depending on the absorbance maxima of the applied substances at 320 to 370 nm.

Quantification of flavonoids by TLC.

To quantify flavonoids in biotransformation and biocatalytic reactions, samples were diluted 1/10 in ethyl acetate-acetic acid (3:1) and subsequently centrifuged. Samples of 20 μl were sprayed by an ATS 4 (Camag, Muttenz, Switzerland) on HPTLC silica gel 60 F₂₅₄ plates (Merck KGaA, Darmstadt, Germany) versus different amounts of respective standard educt and product substances. TLC plates were developed, dried, derivatized, and analyzed as stated above. Regression curves were calculated from the peak area of the applied reference substances to determine the amounts of produced and residual flavonoids.

HPLC-ESI-MS analysis.

HPLC was carried out on a Purospher Star RP-18e 125-4 column (particle size of 3 μm; Merck, Darmstadt, Germany) with a Rheos 2000 pump (Flux Instruments, Suisse) and set pressure limits, with a minimum of 0 Pa and a maximum of 400 × 10⁵ Pa. Injection volumes of 10 μl were separated with solvent A (water supplemented with 0.1% trifluoroacetic acid [TFA]) and solvent B (acetonitrile with 0.1% TFA) under the following HPLC gradient conditions: from 0 min, 0.6 ml/min of 90% A and 10% B; from 14 min, 0.6 ml/min of 75% A and 25% B; from 18 min, 0.6 ml/min of 5% A and B = 95%; from 22 min, 0.6 ml/min of 5% A and 95% B; from 22.1 min, 0.6 ml/min of 90% A and 10% B; and from 28.1 min, 0.6 ml/min of 90% A and 10% B. Elution was monitored with a Finnigan Surveyor photodiode array (PDA) detector, and fractions were collected by an HTC PAL autosampler (CTC Analytics). Mass spectrometry (MS) was performed on a Thermo LCQ Deca XP Plus with an electrospray ionization (ESI) interface in positive ionization.

Sequence analysis.

Automated DNA sequencing of small inserted plasmids was performed using an ABI377 instrument and dye terminator chemistry according to the manufacturer's instructions. Large fosmid sequences were established by 454 sequencing technology. The sequences were assembled by using Gap, version 4, software. ORF finding was performed with Clone Manager Professional, version 9, software.

Nucleotide sequence accession numbers.

All sequences mentioned in this work were deposited in GenBank. The DNA sequence of the Bacillus sp. HH1500 16S rRNA gene was deposited in GenBank under accession number KC145729. The fosmid-derived genes from Bacillus sp. HH1500 identified on subclone pSK4B2 are bspA (JX157885), mgtB (JX157886), and bspC (JX157887), and their sequences have been deposited under accession numbers AGH18135 to AGH18137, respectively. The Elbe sediment metagenome-derived fosmid subclone pSK144C11 comprised genes esmA (JX157626), gtfC (JX157627), esmB (JX157628), and esmC (JX157629). The sequences of the deduced proteins have been deposited under GenBank accession numbers AGH18138 to AGH18141.

RESULTS

Screening method: setup of a TLC-based screening method for the detection of flavonoid-modifying enzyme clones.

Since it is known that Bacillus cereus and Bacillus subtilis encode glycosyltransferases mediating the glucosylation of flavonoids (36), we initially tested several single bacterial isolates from our strain collections with respect to their flavonoid-modifying activities. Biotransformations using whole cells of wild-type isolates confirmed the presence of flavonoid-modifying enzymes in one of the strains. This strain was originally isolated from a soil sample of the botanical garden in Hamburg, Germany, and was designated Bacillus sp. HH1500. Sequence analysis of a 16S rRNA gene (GenBank entry KC145729) showed 100% identity to members of the B. cereus group (data not shown). In order to use this strain as a positive control, we constructed a fosmid library of its genomic DNA in pCC1FOS. The obtained library contained 1,920 clones with an average insert size of 35 kb. Thus, the library encompassed approximately 67 Mb of cloned genomic DNA (gDNA), or about 10 times the average size of a genome from B. cereus group members (43). Further, the sensitivity of the (HP)TLC-based assay was verified using a serial dilution of isoquercitrin, the 3-O-β-d-glucoside of quercetin, by spraying 10 μl of a solution of 0.78 μM up to 100 μM isoquercitrin on TLC plates and measuring the absorbance at 365 nm (see Table S3 in the supplemental material). In addition, 10-μl samples of other glycosylated flavonoids were assayed at 10 μM concentrations and could be detected as clear peaks on the absorbance chromatograms (see Table S3; also data not shown).

Based on the observed sensitivities, we designed a systematic screening scheme, as outlined in Figure 1. Initially 96 fosmid clones were grown in deep-well microtiter plates at 37°C overnight. Cultures were then pooled, and, following this step, the cells were sedimented by centrifugation and resuspended in fresh LB medium containing the appropriate antibiotics and 100 μM quercetin as an acceptor substrate. After incubation for 16, 24, and 48 h at 30°C, 4-ml samples of the pooled cultures were withdrawn and extracted with half the volume of ethyl acetate. Of these extracts 20 μl was applied on TLC silica plates and separated using Universal Pflanzenlaufmittel as a solvent. The absorbance of the developed sample lanes was determined densitometrically at 365 nm. Additionally, bands of substrates and modified flavonoids were visualized by staining with Naturstoff reagent A as outlined in the Materials and Methods section. In our hands the sensitivity of the assay was high enough to detect a single flavonoid-modifying enzyme clone in a mixture of 96 clones (Fig. 1). After the detection of a positive signal, we divided the 96 fosmid clones into pools of 48 to locate the same peak in one of the resulting two microtiter half-plates. Following this procedure, we divided the 48 clones into six groups of eight clones (Fig. 1) and finally analyzed the eight individual clones. This strategy was applied successfully to identify six overlapping positive clones in the Bacillus sp. HH1500 fosmid library testing all 20 microtiter plates with a total of 1,920 clones.

Fig 1.

Outline of the metagenome screening for flavonoid-modifying clones. A schematic workflow of the function-based screening procedure for the systematic identification of flavonoid-modifying clones is shown. Initially, pools of 96 clones preincubated in deep-well plates were tested in biotransformation reactions with quercetin, followed by culture extract TLC analysis as described in Materials and Methods. Once a positive pool forming a product peak was observed, the respective plate was divided in two half-plate pools of 48 clones each, which were tested similarly. The positive fraction was divided into six rows of eight clones per column (or, vice versa, 8 by 6 per row) that were tested for activity until candidate single clones were tested and the positive clones were identified. Putative positive clones were then reconfirmed and subcloned for further analysis. UV chromatograms from TLC analysis are displayed in relative absorbance units (AU) versus the R_f value measured at 365 nm in a densitometric TLC Scanner 3 (Camag, Muttenz, Switzerland). Light gray peaks indicate absorbance of the remaining quercetin (Q) substrate, dark gray peaks (P1) indicate the formed product isoquercitrin. Depicted chromatograms show UV chromatograms of culture extracts in EtOAc after 24-h biotransformations of pools with 96, 48, and 8 clones including fosmid clone pFOS19G2 and of the single clone pFOS19G2 (last).

Of these six fosmid clones, one clone, pFOS4B2, of approximately 46 kb was subcloned using the HindIII restriction site of the pBluescript II SK+ vector. The obtained subclones were analyzed using the above-mentioned TLC screening technology. Thereby, a positive subclone designated pSK4B2 was identified and completely sequenced (GenBank entries JX157885 to JX157887). Subclone pSK4B2 carried an insert of 3,225 bp (see Fig. S1A in the supplemental material) and harbored a gene, designated mgtB, encoding a protein of 402 amino acids (aa). The identified ORF was subcloned, creating plasmid pDmgtB, and again assayed for activity. TLC analysis clearly confirmed the glycosylation activity of the MgtB enzyme in this construct as well. The deduced amino acid sequence of MgtB (GenBank accession number AGH18136) was highly similar to a predicted B. thuringiensis macroside glycosyltransferase (Table 1). The mgtB-surrounding DNA sequences in plasmid pSK4B2 represented two truncated genes that consistently were almost identical to genes from B. thuringiensis (Table 1). This phylogenetic relation was in accordance to the preliminary sequence analysis of the 16S rRNA gene of Bacillus sp. HH1500 (see above).

Table 1.

ORFs identified on subclones pSK4B2, derived from the active Bacillus sp. HH1500 fosmid clone, and pSK144C11, derived from the river Elbe sediment active fosmid clone

Subclone and ORF	Position (aa)	Homolog (accession no.)	Coverage (%)	% identity	% similarity
pSK4B2
bspA	221	Putative protein kinase from Bacillus thuringiensis (EEM66464)	100	99	99
mgtB	402	Macrolide glycosyltransferase from Bacillus thuringiensis (EEM96628)	100	98	99
mgtC	261	Hypothetical membrane protein from Bacillus thuringiensis (EAO54527)	100	99	100
pSK144C11
esmA	80	Putative UDP-N-acetylmuramate–l-alanine ligase from Niabella soli (EHP51575)	99	69	80
gtfC	459	Putative UDP-glucosyltransferase from Fibrisoma limi (CCH52088)	92	51	71
esmB	170	Hypothetical protein from Niastella koreensis (YP005009630)	95	63	77
esmC	150	Putative membrane protein from Solitalea canadensis (YP006258217)	98	68	81

These tests suggested that the screening procedure was suitable for the functional screening of large-insert metagenome libraries. For the function-based screening of metagenomes, we termed this methodology META, for metagenome extract TLC analysis. Although it is not a fully automated high-throughput screening (HTS) technology, META allows screening of about 1,200 clones per TLC plate within a time of 48 h for preculture, biotransformation, and analysis. This number of clones appeared to be feasible if a single person did the screening. Generally, the sampling of about one TLC plate per hour by the ATS 4 is the time-limiting step of the method. But this still allows the pooled screening of several plates a day and, hence, throughput of several thousand clones a day by META.

Identification of a novel glycosyltransferase from a metagenome library.

To further apply the screening for enzyme discovery in metagenome libraries, we tested two fosmid libraries constructed in our laboratory. One library was constructed from DNA isolated from river Elbe sediment; the other was from DNA isolated from fresh elephant feces. Altogether both libraries encompassed approximately 50,000 clones with an average insert size of 35 kb. Both libraries were screened using quercetin as a substrate. Using the described strategy, we discovered one positive microtiter plate pool in the river Elbe sediment library. Further screening of this pool resulted in the identification of a single positive fosmid clone, designated pFOS144C11 (Fig. 2). Biotransformations of quercetin (Fig. 2A, Q) with 48 clone pools presented one product peak (P2) by TLC separation with an R_f value comparable to that of quercitrin, the quercetin-3-O-β-l-rhamnoside (Fig. 2A). A second peak (P3) with an R_f value higher than values for the available reference quercetin glycones was observed in conversions with the six-clone pool and the single fosmid clone (Fig. 2B and C, respectively). Clone pFOS144C11 carried a fosmid of approximately 40 kb. Subsequent restriction fragment subcloning into pBluescript II SK+ with HindIII yielded the identification of the positive E. coli DH5α subclone pSK144C11. However, biotransformations with pSK144C11 showed two product peaks, a major one (Fig. 2D, P2) with an R_f value comparable to that of quercitrin and a minor one (P1) similar to isoquercitrin (Fig. 2D). The subclone pSK144C11 still had an insert of approximately 8.5 kb in size. Further sequencing and subcloning of pSK144C11 finally identified the gene putatively responsible for the modifications, which we designated gtfC (see Fig. S1B in the supplemental material). The deduced 459-aa sequence of the corresponding enzyme revealed motif similarities to UDP-glucuronosyltransferase/UDP-glucosyltransferases. GtfC (GenBank entry AGH18139) showed a similarity of 71% to the putative glycosyltransferase of the Gram-negative bacterium Fibrisoma limi covering 92% of the protein (Table 1). Further cloning of the gtfC ORF into the pDrive vector and biotransformation with E. coli DH5α carrying the respective construct pDgtfC confirmed the flavonoid-modifying activity of GtfC (Fig. 2E).

Fig 2.

Iterative TLC analyses of culture extracts for the identification of a flavonoid-modifying enzyme. Cells were grown in LB medium with appropriate antibiotics. Biotransformation was performed as described in Materials and Methods. Culture extracts in EtOAc from 24-h biotransformation reactions with 100 μM quercetin as a substrate were applied on Merck silica gel 60 F₂₅₄ TLC plates. UV chromatograms are displayed in relative absorbance units (AU) versus the R_f value measured at 365 nm on a densitometric TLC Scanner 3 (Camag, Muttenz, Switzerland) for activity determination. Peaks of the remaining quercetin substrate are depicted in light gray near the solvent front (Q); product peaks are shown in dark gray (P1, P2, and P3). TLC analyses of culture extracts from the following biotransformations led to the final isolation of the GT-encoding ORF gtfC: pool MT144R of 48 fosmid clones within the positive clone pFOS144C11 (A), pool MT144C of six fosmid clones within the positive clone (B), the positive single fosmid clone pFOS144C11 (C), the positive subclone pSK144C11 derived from pFOS144C11 (D), and the active ORF gtfC derived from pFOS144C11 in clone pDgtfC (E).

In summary, these results demonstrated that the developed screening procedure, META, is sufficiently sensitive to allow the identification of large-insert clones from individual bacterial genomes (i.e., Bacillus sp. HH1500) and complex metagenome libraries (i.e., the river Elbe sediment library) showing flavonoid-modifying activities.

Sequence-based classification of MgtB and GtfC.

To analyze the affiliation of MgtB and GtfC, we constructed a phylogenetic tree using the MEGA, version 5, software (44). The amino acid sequences of MgtB and GtfC and their closest sequence-based relatives determined by pBLAST were aligned by ClustalW. Additionally, the sequences of the actually published prokaryotic flavonoid-active GTs were aligned along with those of two eukaryotic enzymes, the flavonoid glucosyltransferase UGT85H2 from Medicago truncatula and the flavonoid rhamnosyltransferase UGT78D1 from Arabidopsis thaliana as an outer group (45, 46). From these sequences, a neighbor-joining tree with 100 bootstraps was computed (Fig. 3). As expected, MgtB from Bacillus sp. HH1500 clustered with other MGTs from the B. cereus group. At the time of writing, the MGT of B. thuringiensis IBL 200 and the MGT of B. cereus G9842 turned out to be the closest relatives, with an amino acid identity to MgtB of 98% each. Both MGTs were annotated as predicted enzymes, and no substrate data were available. From the MGT cluster, five other enzymes were previously reported to mediate the glucosylation of flavonoids. Three of them, BcGT-1, the nearest relative reported to be flavonoid active, BcGT-4, and BcGT-3, all originated from B. cereus ATCC 10987 (47–49). Another flavonoid-active MGT, designated BsGT-3, originates from B. subtilis strain 168 (36). The remaining flavonoid-active MGT is the well-studied OleD from Streptomyces antibioticus (50, 51). GtfC was located in a distinct cluster of UGTs and appeared to be somewhat related to hypothetical enzymes from Cytophagaceae bacteria such as Dyadobacter fermentans and Fibrisoma limi (Fig. 3). Within this cluster only the UGT XcGT-2 is known to accept flavonoid substrates (38). Interestingly, rhamnosyltransferases like BSIG 4748 from Bacteroides sp. strain 116 and RtfA from Mycobacterium avium phylogenetically also show affiliation to this cluster but form a separate branch (Fig. 3).

Fig 3.

Phylogenetic dendrogram of glycosyltransferases (GTs) related to the two GTs (MgtB and GtfC in black boxes) identified in this study. GTs known to act on flavonoids as acceptors are highlighted in gray. Phylogenetic analysis was conducted using MEGA, version 5 (44), with ClustalW sequence alignment in a BLOSUM protein weight matrix. The neighbor-joining tree was calculated using the bootstrap method with the Poisson model; bootstrap values higher than 75 are indicated next to the branches. The scale represents the number of amino acid changes per residue. All GTs shown belong to the family of Gtf-like GT1 and are further subclassified into MGTs (macroside glycosyltransferases; TIGR01426), NGTs (N-glycosyltransferases), and UGT (a UDP-glucuronosyltransferase/-glucosyltransferase; PF00201) as indicated on the right. GenBank numbers were retrieved from NCBI as follows: YP002445489, MGT from B. cereus G9842; ZP04071678, MGT from B. thuringiensis IBL 200; AGH18136, MgtB from Bacillus sp. HH1500; AAP25969 Bacillus anthracis strain Ames; ABY43166, MGT from Bacillus weihenstephanensis KBAB4; NP978481, BcGT-1 from B. cereus ATCC 10987; NP979441, BcGT-4 from B. cereus ATCC 10987; AAS41737, BcGT-3 from B. cereus ATCC 10987; NP389104, BsGT-3 from B. subtilis strain 168; ABA42119, OleD from S. antibioticus; AAM41712, XcGT-2 from X. campestris ATCC 33913; YP003086330, Dfer 1940 from Dyadobacter fermentans DSM 18053; AGH18139, GtfC from Elbe river sediment metagenome; CCH52088, UGT from Fibrisoma limi BUZ 3; YP003388759, Slin3970 from Spirosoma linguale DSM 74; ACV78946, Namu2594 from Nakamurella multipartita DSM 44233; YP712191, FRAAL1959 from Frankia alni ACN14a; AAN01207, RebG from Lechevalieria aerocolonigenes; ZP09941874, BSIG 4748 from Bacteroides sp. 116; AAC71702, RtfA from Mycobacterium avium; AFK05536, Emtol 0266 from Emticicia oligotrophica DSM 17448; 2PQ6A, UGT85H2 from Medicago truncatula; and NP564357, UGT78D1 from Arabidopsis thaliana.

To further characterize the identified metagenome-derived GTs, the amino acid residues of the C-terminal donor binding regions were compared to the motifs of the closest relatives and the known flavonoid-active GTs (Fig. 4). Here, the Rossmann fold α/β/α subdomain, the conserved donor-binding region of UGTs, is located (52). Plant UDP-glycosyltransferases like UGT85H2 and UGT78D1 exhibit a highly conserved motif in this region, which is termed the plant secondary product glycosyltransferase (PSPG) motif (45, 46, 53, 54). By alignment we were able to identify key amino acids known to be of importance for NDP-sugar binding. While MgtB revealed a clear UDP-hexose binding motif consisting of highly conserved Gln289 and Glu310 residues for ribose binding and a conserved DQ, GtfC lacked this motif (45, 55, 56). Instead, GtfC presented typical residues Phe336 and Leu357 for deoxyribose nucleotide utilization (57). Moreover, we were able to identify the pyrophosphate binding sites in the MgtB amino acid sequence (Fig. 4). However, GtfC does not possess these conserved phosphate binding residues, suggesting that GtfC and related enzymes have another donor binding mode. In this context GtfC seemed to belong to a novel enzyme class, as underlined by the low level of sequence homology.

Fig 4.

ClustalW alignment of the MgtB and GtfC (black boxes) amino acid sequences, their nearest sequence-based relatives, and other flavonoid-active GTs (gray boxes). The C-terminal region of the Rossmann fold α/β/α subdomain, the conserved donor-binding region of UGTs, is shown. Plant UDP-glycosyltransferases like UGT85H2 (2PQ6A) from Medicago truncatula and UGT78D1 (NP564357) from Arabidopsis thaliana exhibit the highly conserved plant secondary product glycosyltransferase (PSPG) motif in this region. Amino acids are boxed according to their roles in donor nucleoside (blue), phosphate group (green), and hexose (red) binding, as reported in the literature (45, 52, 55–57, 73, 74). Dark and light green boxes refer to beta- and alpha-phosphate group binding of the donor molecule, respectively. Analogous binding of the nucleoside part is indicated for base positions (light and full bright blue) and ribose positions (dark blue); for the latter, lighter variants of blue boxes differentiate deoxyribose-specific and ribose-specific positions. GenBank numbers were retrieved from NCBI (see the legend of Fig. 3 for species identifications).

Overexpression and glycosylation patterns of MgtB and GtfC.

To further characterize the novel enzymes and verify their functions, we overexpressed and purified MgtB and GtfC as His-tagged proteins in E. coli BL21(DE3). Both genes, mgtB and gtfC, were ligated into the expression vector pET19b. The recombinant enzymes containing N-terminal His₁₀ tags were purified by Ni affinity chromatography under native conditions and gradient elution. MgtB could be purified to yield more than 5 mg of protein/g of cell pellet (wet weight). The maximum yield of GtfC was 3 mg of protein/g of cell pellet. The molecular weights of the proteins were verified by SDS-PAGE analysis under denaturing conditions according to Laemmli. After Coomassie staining, His₁₀-MgtB was visible as a single band with a molecular mass of approximately 50 kDa on a 12% SDS-PAGE gel (see Fig. S2A in the supplemental material). This was in accordance with the calculated molecular mass of 51.2 kDa including the N-terminal His tag. His₁₀-GtfC revealed a molecular mass of about 55 kDa on a 12% SDS-PAGE gel, which was well in accordance with the calculated molecular mass of 54.7 kDa including the N-terminal His tag. While virtually no additional bands were visible on SDS-PAGE gels with purified recombinant MgtB protein, some minor contaminating bands were still visible on the SDS-PAGE gel loaded with purified GtfC (see Fig. S2B in the supplemental material). In summary, both proteins could be purified to allow further biochemical characterization.

The purified His₁₀-MgtB protein was able to use UDP-α-d-glucose as a donor substrate. The recombinant enzyme catalyzed the transfer of α-d-glucose residues to various polyphenols. Biocatalytic reactions were performed with 500 μM UDP-α-d-glucose as a donor and 100 μM acceptor substrate. The following flavonoids served as acceptor substrates and were modified with high yields: luteolin, quercetin, kaempferol, tiliroside, naringenin, and genistein (Table 2). Flavonols turned out to be the best acceptor molecules. Generally, the conversion during a 2-h assay ranged from 52% for naringenin to approximately 100% for quercetin and kaempferol. Interestingly, in the presence of quercetin and kaempferol, no residual educts could be monitored by HPTLC analysis. The specific educts and their observed glycones of the biocatalytic reactions are summarized in Table 2 together with the respective R_f values. MgtB favored the glucosylation at the C-3 hydroxy group, if accessible, as in the aglycone flavonols quercetin and kaempferol. Further, the C-7-OH was attacked and glucosylated by the enzyme, which could be shown for not only the flavone luteolin but also the flavanone naringenin and the isoflavone genistein (Table 2). MgtB glucosylated luteolin also at the C-3′ hydroxy group forming the 3′,7-di-O-glucoside of luteolin if the C-7-OH was glucosylated previously. Furthermore, MgtB catalyzed catalyzed the conversion of the kaempferol derivative tiliroside [kaempferol-3-(O-6″trans-p-coumaryl)-glucoside]. One glucosylated product with an R_f value of 0.54 was detected.

Table 2.

Flavonoid substrates converted by recombinant MgtB in bioassays

^a Reactions were carried out at 37°C for 2 h in triplicate using 1 ml of reaction mixture consisting of 500 μM UDP-glucose, a 100 μM concentration of the respective flavonoid, and 5 μg/ml of purified and recombinant MgtB.

^b R_f values and products in bold indicate the main products of the biocatalytic reactions.

^c Products were not specified due to the lack of available reference substances.

Finally, the chalcone xanthohumol and the stilbene t-resveratrol were tested in biotransformation reactions with E. coli expressing mgtB, but conversions were not quantified (data not shown). Xanthohumol yielded three detectable products, whereas the biotransformation of t-resveratrol yielded one observed product by absorbance TLC analysis.

Tests with recombinant and purified GtfC using UDP-α-d-glucose and UDP-α-d-galactose and quercetin as an acceptor molecule suggested that dTDP-activated sugar moieties were transferred by this enzyme. This finding was confirmed by HPLC-ESI-MS analyses of biotransformation assays (see the following paragraph). Unfortunately, deoxyribose nucleotide-activated hexoses, e.g., dTDP-rhamnoside, were commercially not available to further analyze the obtained reaction products in more detail (58).

Biotransformations with the E. coli strain expressing GtfC and using various polyphenols as substrates yielded conversions ranging from 52% for xanthohumol up to almost 100% turnover for most flavonols tested (Table 3). Quercetin was transformed almost completely after 4-h biotransformations and yielded three detectable products (P1 to P3). To further characterize these products, UV absorbance spectra were recorded and compared to the reference glycones of quercetin isoquercitrin and quercitrin (59). P1 revealed an R_f value identical to the value of isoquercitrin. Further, the UV absorbance spectrum of P1 matched the spectrum of isoquercitrin (see Fig. S3A in the supplemental material). P2 revealed an R_f value identical to the one known for quercitrin. P2 also exhibited the same UV absorbance spectrum as quercitrin (see Fig. S3B). P3 revealed an R_f value of 0.82, which clearly differed from the R_f values of known and available quercetin glycones. Compared to isoquercitrin, P3 showed a similar hypsochromic shift of band I to a maximum wavelength (λ_max) of 363 nm (see Fig. S3C); however, it revealed a less hypsochromic shift in band II of only 5 nm to 272 nm with a shoulder at 280 nm. It is further notable that the HPLC-ESI-MS analysis of biotransformation products of quercetin consistently identified three distinct reaction products (see Fig. S4). P1 had a retention time (RT) of 17.93 min in the HPLC analysis and revealed a molecular mass of 464 Da, which is equivalent to isoquercitrin. P2 revealed an RT of 18.06 min and had a molecular mass of 448 Da. This mass corresponds well with the molecular mass of quercitrin. Finally, P3 with a RT of 18.31 min revealed a molecular mass of 446 Da, indicating the formation of a novel, not further characterized, quercetin glycoside.

Table 3.

Flavonoid substrates and products of biotransformation assays with recombinant GtfC

^a Quantification of the reaction products was performed as stated in Materials and Methods. Triplicate reactions using 50 ml of reaction mixture were performed in 50 mM sodium phosphate buffer, pH 7.0, containing 1% (wt/vol) glucose and 200 μM flavonoid at 30°C.

^b R_f values and products in bold indicate the main products of the biotransformation reactions.

^c Products were not specified due to the lack of available reference substances.

Glycosylation patterns of GtfC on quercetin suggested a preference to act on the C-3 hydroxy group mediating the transfer of different sugar residues. However, if a C-3-OH group was not available, GtfC efficiently catalyzed the glycosylation of other positions. Flavones lacking the hydroxy function at C-3 were converted depending on the availability of other hydroxy groups. Pratol, which possesses only a single free C-7 hydroxy group, was converted weakly and yielded a single detectable product. Further, the biotransformation of 3′,4′-dihydroxyflavone yielded three detectable glycones, and 5-methoxy-eupatorin yielded two products (see Table S4 in the supplemental material; also data not shown); the biotransformation of the mono-4′-hydroxyflavanone yielded one glycosylated product, and the glycosylation of naringenin yielded two products. The major biotransformation product of naringenin revealed the same R_f values and absorbance spectra as prunin, the naringenin-7-O-glucoside (Table 3). The second naringenin glycone could not be further specified due to the lack of commercially available reference substances. Altogether these results suggested that GtfC acts on the C-3, C-3′, C-4′, and C-7 hydroxy groups of the flavonoid backbone.

In summary these data demonstrated that MgtB and GtfC possess interesting biocatalytic properties. While MgtB specifically mediated the transfer of glucose residues, GtfC transferred different hexose moieties. MgtB was capable of catalyzing the glucosylation of already glycosylated flavonoids to form diglycosides (e.g., formation of luteolin-3′,7-di-O-glucoside) and even tiliroside to generate novel glucosides not available from natural resources. In contrast, the glycosylation pattern of GtfC suggested the transfer of single sugar residues to only aglycone flavonoid forms. Interestingly, GtfC seemed to be very variable with respect to its activity at various positions on the flavonoid backbone. This may lead to the formation of truly novel flavonoids not available naturally. Hence, both enzymes might be helpful in the generation of new natural compounds.

DISCUSSION

Within the manuscript, we report on the development of a semiautomated TLC-based detection system for flavonoid-modifying enzyme clones. The screening assay was highly reproducible and highly sensitive. It allowed the detection of micromolar concentrations of glycosylated flavonoids. Isoquercitrin was detectable at a 0.78 μM concentration in the assay (see Table S3 in the supplemental material). Using this assay we were able to systematically identify one positive clone out of pools of 96 metagenome clones (Fig. 1 and 2). To our knowledge this is the first published TLC-based screening method for functional searches in metagenome libraries. We speculate that slight modifications of this screening system could easily allow the detection of other flavonoid-modifying enzyme clones, for example, through acylation or methylation reactions.

Using this novel screening technology, we identified a macroside glycosyltransferase, MgtB, from a soil isolate (i.e., Bacillus sp. HH1500). A fosmid library established with DNA from this strain, which had been isolated from the local botanical garden only recently, was initially used to develop and verify the outlined screening technology; and using the novel screening technology, MgtB was quickly identified from a pool of almost 2,000 clones. Isolation and purification of recombinant MgtB revealed a novel MGT. MgtB shared 89% amino acid identity with BcGT-1 from B. cereus ATCC 10987, the closest relative reported to act on flavonoids. BcGT-1 was reported to catalyze the glucosylation of flavones, flavonols, flavanones, and isoflavones (47). On flavonols BcGT-1 acted on C-3, C-7, and C-4′ hydroxy groups creating triglucosides of kaempferol (48). In contrast, biocatalysis of kaempferol with MgtB yielded just two detectable glucosylated products. However, reactions with quercetin resulted in three detectable glycones. These data suggested that MgtB acted at the C-3′-OH group. This hypothesis also was supported by the observation that recombinant MgtB converted luteolin to luteolin-3′,7-di-O-glucoside as a by-product. These results were in accordance with the glucosylation pattern of BcGT-3, yet another MGT from B. cereus ATCC 10987 (49). Interestingly, BcGT-3 shares only 40% amino acid identity with MgtB, but both enzymes act on the same flavonoids, forming diglucosides from flavones and flavonols at the same positions and only monoglucosides from naringenin. The most spectacular conversion observed for MgtB was that of tiliroside. The product is likely to be the 7-O-glucoside, taking the glycosylation pattern of MgtB into account. Tiliroside glycosides, however, have not been reported in scientific literature. This raises the possibility of the generation of new natural compounds. The natural substrates of Bacillus MGTs still have not been reported. Other MGTs like OleD usually detoxify macroside antibiotics but often possess broad acceptor tolerance (35, 60).

The metagenome-derived GtfC turned out to be a completely novel enzyme. Only seven flavonoid-active UGTs have been reported so far that originate from five different prokaryotes (35, 36, 38, 47, 49). With the exception of XcGT-2 from the Gram-negative X. campestris ATCC 33913, all are MGT enzymes from Gram-positive Bacilli and Streptomycetes. MGTs play an important role in xenobiotic defense mechanisms of prokaryotes and thus show broad acceptor specificities (55, 60). This also applies to eukaryotic UGTs, pointing to a biological principle of detoxification (61). To our knowledge GtfC is the first metagenome-derived GT acting on flavonoids. Moreover, it is also the first bacterial enzyme reported to transfer various dTDP-activated hexose sugars to polyphenols (see below), in contrast to the usually stringent donor specificities of Gtf-like enzymes such as GtfD (57). With respect to the notion that many NDP-sugars in prokaryotes are dTDP and not UDP activated, GtfC might be a promising biocatalyst in glyco-diversification approaches (58, 62, 63). GtfC is similar to predicted GTs from Cytophagaceae bacteria (64–66). These Gram-negative bacteria have large genomes, suggesting extensive secondary metabolic pathways, and they are well known for the presence of resistance mechanisms to antibiotics such as trimethoprim and vancomycin (67, 68). As commonly known, glycosylation of xenobiotics is a ubiquitous detoxification process in all kingdoms of life. The phylogenetically diverse members of Cytophagaceae have only recently become an object of research, and concrete estimation of the phylogenetic breadth of this family and exact taxonomic ranking still remain unclear (65, 69). Thus, the identification of the metagenome-derived GtfC and its partial characterization suggest that this group of microorganisms is perhaps a highly promising resource for novel GTs and also other enzymes.

A ClustalW alignment of the donor-binding region of GtfC suggested that the activated donor substrates are of deoxy-thymidine nucleoside origin. GtfC possesses the typical amino acid residues Phe336 for thymine base stacking and hydrophobic Leu357 for deoxyribose fitting (57). Concerning the donor binding of GTs, GtfC appears not to exhibit the known amino acid residues for pyrophosphate binding (Fig. 4). Instead of the conserved residue His/Arg in the current solved protein structures, GtfC contains an Asn at amino acid position 349 (52, 70). This applies also for the nearest GtfC relatives Dfer1940, UGT of F. limi BUZ 3, and Slin3970 as well as the NGTs RebG and BSIG4748. Further, GtfC does not show the conserved Ser/Thr residue responsible for α-phosphate binding. Instead, the Gly354 appears to be of importance for the α-phosphate binding, similar to structure of the OleD transferase (55).

The assumption of dTDP-activated cosubstrates used by GtfC was supported by the observation that glucose, rhamnose, and a third sugar residue with a molecular weight of 446 were transferred by GtfC in biotransformations using intact E. coli cells. Moreover, biocatalytic approaches with purified GtfC and either UDP-α-d-glucose or -galactose as a donor substrate failed. In bacteria, the activated sugars, dTDP-α-d-glucose, -4-keto-6-deoxy-α-d-glucose or -4-keto-β-l-rhamnose, and -β-l-rhamnose, are part of the dTDP-sugar biosynthesis pathway and are present in E. coli (71). Moreover, levels of dTDP-sugars are allosterically regulated by dTDP-rhamnose levels through activity of RmlA (72).

In summary, the screening protocol described in this report is a very helpful tool for the identification of truly novel enzymes for the modifications of flavonoids and related substrates. In this study, we have used the technology to identify two novel flavonoid-modifying enzymes. Both of these enzymes would perhaps not have been detected without the above-developed screening technology. The partial biochemical characterizations using either biocatalysis or biotransformation suggest that MgtB and GtfC are both very interesting enzymes with high potential for biotechnological applications with respect to flavonoid modifications on an industrial scale. Thus, future work will now refine this technology to also identify other enzymes linked to flavonoid modifications.