Functional metagenomics of the thioredoxin superfamily.

Environmental sequence data of microbial communities now makes up the majority of public genomic information. The assignment of a function to sequences from these metagenomic sources is challenging because organisms associated with the data are often uncharacterized and not cultivable. To overcome these challenges, we created a rationally designed expression library of metagenomic proteins covering the sequence space of the thioredoxin superfamily. This library of 100 individual proteins represents more than 22,000 thioredoxins found in the Global Ocean Sampling data set. We screened this library for the functional rescue of Escherichia coli mutants lacking the thioredoxin-type reductase (ΔtrxA), isomerase (ΔdsbC), or oxidase (ΔdsbA). We were able to assign functions to more than a quarter of our representative proteins. The in vivo function of a given representative could not be predicted by phylogenetic relation but did correlate with the predicted isoelectric surface potential of the protein. Selected proteins were then purified, and we determined their activity using a standard insulin reduction assay and measured their redox potential. An unexpected gel shift of protein E5 during the redox potential determination revealed a redox cycle distinct from that of typical thioredoxin-superfamily oxidoreductases. Instead of the intramolecular disulfide bond formation typical for thioredoxins, this protein forms an intermolecular disulfide between the attacking cysteines of two separate subunits during its catalytic cycle. Our functional metagenomic approach proved not only useful to assign in vivo functions to representatives of thousands of proteins but also uncovered a novel reaction mechanism in a seemingly well-known protein superfamily.

A combined bioinformatics and functional metagenomics approach to discovering lipolytic biocatalysts.

The majority of protein sequence data published today is of metagenomic origin. However, our ability to assign functions to these sequences is often hampered by our general inability to cultivate the larger part of microbial species and the sheer amount of sequence data generated in these projects. Here we present a combination of bioinformatics, synthetic biology, and Escherichia coli genetics to discover biocatalysts in metagenomic datasets. We created a subset of the Global Ocean Sampling dataset, the largest metagenomic project published to date, by removing all proteins that matched Hidden Markov Models of known protein families from PFAM and TIGRFAM with high confidence (E-value > 10(-5)). This essentially left us with proteins with low or no homology to known protein families, still encompassing ~1.7 million different sequences. In this subset, we then identified protein families de novo with a Markov clustering algorithm. For each protein family, we defined a single representative based on its phylogenetic relationship to all other members in that family. This reduced the dataset to ~17,000 representatives of protein families with more than 10 members. Based on conserved regions typical for lipases and esterases, we selected a representative gene from a family of 27 members for synthesis. This protein, when expressed in E. coli, showed lipolytic activity toward para-nitrophenyl (pNP) esters. The K m-value of the enzyme was 66.68 µM for pNP-butyrate and 68.08 µM for pNP-palmitate with k cat/K m values at 3.4 × 10(6) and 6.6 × 10(5) M(-1)s(-1), respectively. Hydrolysis of model substrates showed enantiopreference for the R-form. Reactions yielded 43 and 61% enantiomeric excess of products with ibuprofen methyl ester and 2-phenylpropanoic acid ethyl ester, respectively. The enzyme retains 50% of its maximum activity at temperatures as low as 10°C, its activity is enhanced in artificial seawater and buffers with higher salt concentrations with an optimum osmolarity of 3,890 mosmol/l.