Phylogenetic screening of a bacterial, metagenomic library using homing endonuclease restriction and marker insertion.

Metagenomics provides access to the uncultured majority of the microbial world. The approaches employed in this field have, however, had limited success in linking functional genes to the taxonomic or phylogenetic origin of the organism they belong to. Here we present an efficient strategy to recover environmental DNA fragments that contain phylogenetic marker genes from metagenomic libraries. Our method involves the cleavage of 23S ribsosmal RNA (rRNA) genes within pooled library clones by the homing endonuclease I-CeuI followed by the insertion and selection of an antibiotic resistance cassette. This approach was applied to screen a library of 6500 fosmid clones derived from the microbial community associated with the sponge Cymbastela concentrica. Several fosmid clones were recovered after the screen and detailed phylogenetic and taxonomic assignment based on the rRNA gene showed that they belong to previously unknown organisms. In addition, compositional features of these fosmid clones were used to classify and taxonomically assign a dataset of environmental shotgun sequences. Our approach represents a valuable tool for the analysis of rapidly increasing, environmental DNA sequencing information.

INTRODUCTION

Metagenomics is defined as the study of the total genomic composition of the biota from a particular environment. In recent years metagenomics has revealed many novel phylogenetic and functional properties of microbial systems (1–4) and has provided insight into the microbial communities of seawater (5–7), soil (8,9), biogas plants (10), indoor air handling unit (11) and in animal- or human-associated niches (12–15). One major aim of metagenomics is to define the functional properties of microorganisms, which are yet to be cultured or difficult to isolate, by directly analyzing their genomes from the environmental sample.

Random shotgun sequencing of environmental DNA has established itself as a core methodology in metagenomics and recent advances in high-throughput sequencing technology have made this approach economically feasible and hence widely used (16). Typically, shotgun sequencing data are functionally annotated on the level of individual sequence reads or assembled contigs. The individual read annotation and poor success in assembling complex (i.e. species-rich) samples (17) often results in functional genes being physically detached from genes containing taxonomic information (such as the 16S rRNA gene). This has limited the progress in revealing the phylotype–function relationships for many uncultivated organisms and several computational and laboratory-based approaches have been developed to overcome this situation (18). Computational methods mainly use binning of DNA fragments based on conserved compositional features (such as oligonucletotide or k-mer frequencies) followed by taxonomic classification of bins. Algorithms like PhyloPythia (19), TETRA (20) or TACOA (21) have been successfully employed; however, their performance is heavily dependent on the availability of reference (or training) sequences that are long enough to provide sufficient compositional information to define a classifier for shorter, unknown fragments (typically a few thousand base pairs in length). These reference fragments should ideally also contain well-established and universal markers for detailed phylogenetic analysis and taxonomic assignment. Fosmids and BAC clones containing phylogenetic markers such as the 16S rRNA gene have often been quoted as fitting these requirements for suitable reference sequences (19,20).

Laboratory-based molecular screens for phylogenetic markers from fosmids or BAC clones have been developed and involve PCR- or fluorescent in situ hybridization-based screening of individually arrayed clones or small pools of clones (around 96 individual clones) (22–25). PCR-based screening of metagenomic libraries clones is frequently employed to identify phylogenetic marker genes (26–30). Potential cross-hybridization of the PCR primers or probes with the host background of the library (e.g. Escherichia coli) limits these approaches to either very specific organisms (e.g. targeting defined species) or to groups distantly related to the library host (e.g. screen for archaeal markers in an E. coli host library). Furthermore, these methods are rather labor intensive, potentially quite costly and often reach their practical limits with increasing clone numbers in the library.

Here we present an efficient phylogenetic screening strategy for metagenomic libraries using homing endonuclease restriction and marker insertion (HERMI) (Figure 1). This approach allows rapid selection of clones containing rRNA genes from large pools by employing a restriction digest with the homing endonuclease I-CeuI. The recognition site of I-CeuI is 26-bp long and cuts very specifically in a conserved part of the intron-free 23S rRNA gene to produce a 4-nucleotide (CTAA) 3′ overhang (31). The slow evolutionary rate of the recognition site as well as the tolerance of the enzyme for minor sequence changes means that rRNA genes from a wide range of organisms can usually be cut (32). Hence, applying a I-CeuI digest to pooled metagenomic clones should only linearize clones containing 23S rRNA genes. The linearized clones are subsequently isolated by inserting a selectable marker gene (i.e. antibiotic cassette), followed by retransformation and selection from the library pool (Figure 1).

Figure 1.

Schematic diagram showing the steps involved in the HERMI process.

MATERIALS AND METHODS

Bacterial strains and media

Escherichia coli EPI300 (Epicentre, Madison, WI, USA) and DH5α strains were used throughout this study. Cultures were maintained in Luria Bertani medium plus 10% NaCl (LB10) or on LB10 with 1.5% agar solid media. For clones containing only the pCC1FOS vector backbone the media was supplemented with 12.5 µg/ml chloramphenicol. For pGEM:CeuKan clones 50 µg/ml kanamycin and 100 µg/ml ampicillin and for pCC1FOS:HERMI clones 12.5 µg/ml chloramphenicol and 50 µg/ml kanamycin were used.

Construction of pooled fosmid library

Total DNA was extracted from the microbial community associated with the marine sponge Cymbastela concentrica (Thomas et al., unpublished data). A metagenomic clone library was constructed using the CopyControl™ Fosmid Library Production Kit (Epicentre, Madison, WI, USA) according to the manufacturer’s protocol with minor modifications. Briefly, environmental DNA was size fractionated and DNA with molecular weight of 36-40 kb was gel excised followed by an end-repair reaction. The DNA (∼1 µg) was ligated with the fosmid vector pCC1FOS followed by packaging the DNA into lambda phage particles, which were used to infect E. coli EPI300 cells. Transformants were plated onto LB10 agar with chloramphenicol and grown at 37°C for 16 h. The resulting library contains approximately 6500 clones with average insert size of 36 kb (data not shown). The E. coli library clones were pooled followed by the addition of 5 volumes of fresh LB10 broth supplemented with chloramphenicol. Induction of the fosmids to high copy number was done by addition of 0.01% (w/v) arabinose and incubating the cultures at 37°C for 4 h. After induction the cells were collected and the fosmid DNA was extracted using the Illustra plasmidPrep Mini Spin Kit (GE Bio-Science Corp, NJ, USA) according to manufacturer’s instructions.

Construction of a kanamycin cassette with flanking I-CeuI recognition sequences

The kanamycin cassette with flanking I-CeuI recognition sequences was generated via a three-step PCR as direct PCR with primers containing the 26-bp I-CeuI recognition sequence caused severe primer–dimer formation. The first round of PCR includes the amplification of the kanamycin resistance cassette from plasmid pACYC177 (NEB, Beverly, MA, USA) with flanking partial I-CeuI recognition sequences. The PCR conditions as per 50 µl reaction were 10 ng of pACYC177 plasmid, 1 × iProof HF Buffer, 200 µM of each dNTP, 0.5 μM of each primer (KanFCeu/KanRCeu) (Table 1) and 0.05 U/µl iProof DNA polymerase (Bio-Rad, Hercules, CA, USA), 1 min/98°C, followed by 10s/98°C, 30 s/53°C, 1 min/72°C, 30 cycles. The second round of PCRs was performed in two separate reactions to add the entire I-CeuI restriction sequence on each of the ends of the cassette; one reaction used the primer set CeuF/KanRCeu and the other used primer set CeuR/KanFCeu (Table 1). PCR conditions were: 3 ng of DNA template from the first PCR, remaining reaction mix composition as for the first-round PCR, 1 min/98°C, followed by 10 s/98°C, 30 s/45°C, 1 min 30 s/72°C, 30 cycles. The two PCR products were then pooled (final volume of 100 µl) and the 1.1 kb product was gel extracted to purify the kanamycin cassette. The third round of PCR involved hybridizations and fill-in reaction steps to generate a PCR product with the full I-CeuI recognition site on both ends of the kanamycin cassette. The conditions were: ∼50 µl of gel-extracted DNA from the second round of PCR, 1 × RedTaq PCR buffer, 10 μM of each dNTP and 1 U of RedTaq DNA polymerase (Promega, Madison, WI, USA), 2 min/94°C, 20 s/94°C, 30 s/45°C, 20 s/72°C, 20 cycles. The final PCR product was cloned into pGEM vector (pGEM®-T Easy Vector System, Promega, Madison, WI, USA) according to manufacturer’s instructions followed by the transformation into E. coli DH5α. Transformants (named pGEM:CeuKan) with the correct insert (after confirmation via restriction digest with I-CeuI) were selected for miniprep purification. The kanamycin cassette with flanking I-CeuI recognition sequences (CeuKan) was gel purified after digestion of pGEM:CeuKan.

Table 1.

Primers used in this study

Name		Sequence (5′ to 3′)
CeuF		GGGTAACTATAACGGTCCTAAGGTAGCGAG
CeuR		GGGTCGCTACCTTAGGACCGTTATAGTTAAA
KanFCeu		GTCCTAAGGTAGCGAGGTTGATGAGAGCTTT GTTG
KanRCeu		AGGACCGTTATAGTTAAAGTCAGCGTAATG CTCTGC
KanFSeq		GACCGTTCCGTGGCAAAG
KanRSeq		GCTCGATGAGTTTTTCTAATC
23S–129F		CYGAATGGGRVAACC
23S–2241R		ACCGCCCCAGTHAAACT
16S–27F		AGAGTTTGATCMTGGCTCAG
16S–1492R		ACGGTTACCTTGTTACGACTT
pEpiFosFor		GGATGTGCTGCAAGGCGATTAAGTTGG
pEpiFosRev		CTCGTATGTTGTGTGGAATTGTGAGC
		CCTACGGGAGGCAGCAG
907R		CCGTCAATTCCTTTGAGTTT
GM5FplusGC		CCTACGGGAGGCAGCAGCGCCCGCCGCGCC CCGCGCCCGTCCCGCCGCCCCCGCCCG

Homing endonuclease restriction and marker insertion

One microgram of pooled fosmid DNA was digested with the I-CeuI homing endonuclease in a reaction containing two units of I-CeuI enzyme (NEB), 1 × NEB buffer 4, 0.5 µl of BSA (NEB) and sterile, deionized water up to a volume of 50 µl. The digest was incubated at 37°C for 3 h followed by a heat inactivation at 65°C for 20 min. The digested fosmids were dephosphorylated using two units of Antarctic alkaline phosphatase (NEB) according to manufacturer’s instructions. The fosmids were ligated with the CeuKan cassette in 1:30 (vector:insert) molar ratio overnight at room temperature. The ligation mix was transformed into E. coli EPI300 cells via electroporation and cells were recovered at 37°C for 1 h and then plated onto LB10 agar supplemented with appropriate antibiotics to select for pCC1FOS:HERMI clones. Transformants that grew on the selective agar were purified, fosmid DNA was extracted and subjected to denaturing gradient gel electrophoresis (DGGE) analysis, end sequencing and 16S/23S rRNA gene PCR as described below.

Denaturing gradient gel electrophoresis

The 16S rRNA gene was PCR amplified as described by Muyzer et al. (33) with the universal primer GM5FplusGC and 907RC (Table 1). Various DNA samples were used: (i) Pooled fosmid library DNA, (ii) E. coli genomic DNA and (iii) fosmid DNA of individual HERMI clones. PCR conditions were 10 ng of DNA template, 1 × RedTaq buffer, 0.5 μM of each forward and reverse primers, 200 μM of each dNTP, 300 µg of BSA, 1 U of RedTaq DNA polymerase (Promega, Madison, WI, USA), 3 min/96°C, hot start at 80°C, 30 s/94°C, 30 s/57°C, 1 min 10 s/72°C, 25 cycles. The PCR products were cleaned using the QIA quick PCR purification kit (QIAGEN, Hilden, Germany) and the DNA was examined with a DCode DGGE unit (BIO-RAD, Hercules, CA, USA) using the following parameters: 10% acrylamide gel, a denaturant gradient containing 45–60% urea-formamide, 1× TAE buffer, 75 V at 60°C for 16 h. Bands from the DGGE gel were extracted, dialysed overnight at 4°C with 50 µl of molecular grade water and re-amplified using primers GM5F and 907RC for sequencing.

Sequencing and phylogenetic analysis of fosmid clones

End sequencing of the HERMI clones were performed using the primer pair pEpiFosFor and pEpiFosRev (Table 1). PCR amplification and sequencing using universal primers for 23S and 16S rRNA gene (Table 1) were also performed on selected HERMI clones as described previously (34,35). Briefly, PCR conditions were 3 min/94°C, 1 min/94°C, 1 min/57°C, 3 min/72°C, 30 cycles (for 23S PCR) and 3 min/94°C, 80°C hot start, followed by 30 s/94°C, 1 min/50°C, 3 min/72°C, 25 cycles (for 16S PCR). The PCR products were subjected to sequencing using the same primers. Other sequencing reactions were also performed using the KanFSeq and KanRSeq primers (Table 1) to obtain 23S rRNA gene sequence flanking the kanamycin cassette.

The complete sequence of PCR products were obtained and searched with the BLAST algorithm (36) against the NCBI and Silva database (37) and closest representatives were selected, aligned using the Aligner tool provided in Silva, and imported into the Silva 16S rRNA and 23S rRNA database using the ARB program for phylogenetic tree construction (38). Maximum likelihood trees were constructed with default parameters.

Whole fosmid sequencing and analysis

HERMI fosmid clones were shotgun sequenced as outlined in Rusch et al. (5). Shotgun reads were processed and assembled using Phred/Phrap and the assembly was manually checked in Consed (39). Binning of the fosmid sequences and an assembled shotgun-reads dataset from the bacterial community of C. concentrica (Thomas et al., unpublished data) was done according to the strategy outlined by Woyke et al. (40). Briefly, tetranucleotide patterns were determined using TETRA (20) and exported as normalized Z-scores for fosmids and scaffolds of the shotgun datasets longer than 20 kb. Clustering was performed with Euclidian distance and complete linkage using the software Cluster 3.0 and visualized with JavaTreeView (41).

RESULTS

Specificity of the HERMI screen

Applying the HERMI method in a single-tube reaction to a pooled library of approximately 6500 fosmid clones derived from the microbial community associated with the sponge C. concentrica resulted in 52 kanamycin-resistant transformants. End sequencing of these clones rejected 40 as being sister clones leaving 12 for further analysis. PCR amplification and sequencing of the region flanking the CeuKan cassette showed that none of the clones contains more than one CeuKan cassette. In addition, all insertion sites have similarity to the 23S rRNA gene indicating that no unspecific insertion had taken place. The I-CeuI recognition sequences for four of the HERMI clones that contained unique phylotype (HERMI06, 11, 16 and 30) were compared with the I-CeuI recognition sequence described in (32). This demonstrated that HERMI06, 11 and 30 have perfect matches to the recognition sequence (Figure 2), while HERMI16 had base pair insertions between positions 4–5, 5–6, 6–7 and 7–8 (respective to the original recognition sequence). This indicates that base pair insertions in these positions of the recognition sequence do not abolish I-CeuI cleavage activity.

Figure 2.

Alignment (5′ to 3′) of the I-CeuI recognition sites with insertion sites of HERMI clones and different sequences from the Silva LSU 23S rRNA database. Black triangle represents the cleavage site for I-CeuI and where the CeuKan marker was inserted for the HERMI clones. Letters underlined in the recognition sequence of HERMI16 indicate basepair insertions into the recognition sequence while letters underlined for other sequences highlight mismatches in positions that have previously been shown to negatively impact I-CeuI enzymatic activity.

To evaluate the phylogenetic spectrum of sequences that can be cleaved by I-CeuI, we compared the 26-bp recognition sequence to the entire bacterial 23S rRNA Silva database (37) with various thresholds (no mismatch, 1 mismatch and 2 or more mismatches). Partial sequences (those did not cover the entire region of the I-CeuI recognition sequence), mitochondrial and chloroplast DNA, and sequences belonging to the actinobacteria phyla [previously shown not to be cleaved by I-CeuI (42)] were excluded. We also considered sequences containing mismatches that were previously shown to abolish the activity of the enzyme in our comparison (31). Our analysis showed that only 1.1% of the 23S rRNA gene sequences (59 sequences out of 5525 sequences) contain mismatches that can potentially abolish the enzymatic activity. Phylotypes that might be excluded from our screening approach include a number of species within the genus Chlamydophila and Chlamydia, the unclassified gamma-proteobacteria Candidatus Carsonella ruddii, the Candidate division TM7 group GTL1 and a number of species in the genus Chloroflexi and Roseiflexus (all one mismatch) as well as some species within the Planctomycete, Coxiella burnetii and Francisella tularensis subsp. (two or more mismatches) (Figure 2). This updated, theoretical consideration is consistent with previous observations (43) and given the degree of mismatch tolerance described for homing endonucleases (44,45), we estimate that our HERMI strategy is capable of capturing at least 98% of the known diversity from a wide phylogenetic spectrum of 23S rRNA gene sequences.

Phylogenetic analysis of HERMI clones from C. concentrica

The 12 unique HERMI clones (defined as clones having unique fosmid end sequences) were subjected to 16S and 23S rRNA gene sequencing and their phylogenetic identities were examined. Based on 16S rRNA sequences four unique phylogenetic groups (with a percentage identity cut-off of 99%) were identified from these 12 clones. These four unique 16S rRNA gene sequences (using clone HERMI06, 11, 16 and 30 as representatives) were also found by band sequencing and migration distance to correspond to the four dominant bands of a 16S rRNA gene DGGE analysis of the pooled fosmids. This indicates that we have captured the major phylotypes present in the library (Figure 3). One minor band of the pooled fosmids was found to belong to the group Actinobacteria, which is known not to be cleaved by the I-CeuI enzyme (see above).

Figure 3.

DGGE gel of the 16S rRNA gene of a pooled fosmid library (lane 1), E. coli genomic DNA (lane 2) and HERMI clones (lanes 3–6). Individual bands were excised and sequenced; the taxonomic assignment of the band is shown next to the arrow.

A maximum likelihood tree was constructed for the 16S rRNA sequences of the HERMI clones (Figure 4) and shows that all sequences (except HERMI16, see below) are related to uncultured organisms and at least 7.5% divergent from their nearest cultured bacterium. Taxonomic classification assigned them to novel clades within the α- and γ-proteobacteria group (HERMI06 and HERMI11) and a Mesorhizobium-related group (HERMI30). The 16S rRNA sequence of HERMI16 is identical to E. coli, but further analysis of this fosmid showed that only the 23S rRNA was cloned and that the 16S rRNA sequence resulted from amplification of E. coli host background. A maximum likelihood tree was also constructed for the 23S rRNA gene of the HERMI clones (Figure 5) and their closest representatives imported from the 23S rRNA Silva database (37). Here the HERMI clones are at least 11% divergent to their closest cultured representatives. The longer branch length in this tree to the next related sequences is a reflection of sparseness of available 23S rRNA gene data, but nevertheless allows us to link the 16S rRNA with 23S rRNA gene phylogeny (Figure 5).

Figure 4.

Maximum likelihood tree of the 16S rRNA gene for the HERMI clones (bold). and the closest representatives of cultured (from the RDP database, underlined) and uncultured organisms (from Silva SSU database) with accession numbers preceding the name of the organism. Scale bar indicates 0.025 sequence divergence.

Figure 5.

Maximum likelihood tree of the 23S rRNA gene for the HERMI clones (bold) and the closest representatives from the Silva 23S LSU database. Underlined sequence highlights closest cultured representatives. Scale bar indicates 0.025 sequence divergence.

Binning and taxonomic assignment of shotgun metagenomic sequences

To illustrate the value of whole fosmid sequences containing 16S or 23S rRNA genes for the assignment of unknown DNA fragments, we clustered the whole sequence of HERMI11 with sequences longer than 20 kb from an assembled shotgun sequence dataset of the bacterial community of C. concentrica. As illustrated in Figure 6, the HERMI11 clone sequence fell into a cluster of unknown sequences (red branches and yellow box) with a high correlation index (0.884). We therefore can now assign the shotgun fragments in this cluster to the novel phylotype within the γ-proteobacteria and link this taxon with 372 predicted protein contained within these DNA sequences.

Figure 6.

Clustering of HERMI11 fosmid sequence with assembled shotgun sequences of unknown taxonomic assignment. Arrow indicates the position of HERMI11 in a cluster of sequences (correlation of 0.884) that is highlighted by red branches and a yellow box. Cyan and red colors in the data matrix represent over- and under-representation of tetranucleotide Z-scores.

DISCUSSION

In this study we have shown that HERMI is a suitable tool for screening a metagenomic library for clones with phylogenetic markers (i.e. the 23S rRNA gene). Handling time for the HERMI procedure is less than 2 h and allows for the simultaneous screening of thousands of clones in a single-tube reaction, hence making our approach faster and more convenient than other phylogenetic marker screens (22–25). We also note that the selectable marker applied in this study (i.e. kanamycin resistance marker) could be replaced with any marker cassette. For example, a green fluorescent protein cassette could be employed allowing for subsequent screening of clones with fluorescence-activated cell sorting (FACS).

Nesbø et al. (42) also recently used I-CeuI to clone fragments containing 23S rRNA genes from environmental DNA samples. Their strategy involved the construction of a specialized vector that allowed for the insertion of DNA digested with a blunt-end restriction enzyme and I-CeuI. However, their approach did not yield full-length 23S and 16S rRNA genes, limiting the phylogenetic and taxonomic value of their clones. The HERMI approach does not have this limitation and, as 16S and 23S rRNA genes are often tightly clustered together, will frequently yield clones with 16S rRNA information, for which more comprehensive databases are currently available. We also note that the linkage of 16S and 23S rRNA information will also help to improve annotation of unknown clades in current 23S rRNA databases, as illustrated by the phylogenetic analysis in our study (Figure 4 and 5). Another benefit of the HERMI strategy is that it can be applied retrospectively to any existing environmental DNA library, many of which have so far only been screened for expressed, functional genes (46–48). Screening those libraries for phlyogenetic markers will generate more information about compositional-features of the DNA from uncultured organisms, which in turn will improve our ability to classify bins of environmental DNA sequences (Figure 6) and hence reveal more phylotype–function relationship.

Sequence analysis of the 26-bp recognition site in the Silva 23S rRNA database indicated that only a small proportion (<2%) of all currently known 23S rRNA genes might not be captured with our approach. Importantly, this limitation can be quantified for each sample prior to the screening, for example, by constructing a 23S or 16S rRNA gene PCR library and determine its phylogenetic composition. This is typically done as a first step for most environmental diversity studies (49) and hence would not impose an additional work load.

Finally, homing endonucleases are not only restricted to cleaving the 23S rRNA genes. Arnould and coworkers have recently demonstrated that a semi-rational design can yield I-CreI mutant homing endonucleases with highly specific recognition patterns, which are distinct from the wild-type activity (50). As such other homing endonuclease mutants are available or could be designed that target other common genes used for phylogenetic analysis (e.g. the 16S rRNA gene, recA), functional genes or non-coding regions, and which could then be applied in a analogous screening strategy as described here. This will further enhance the utility and versatility of the HERMI approach.

ACCESSION NUMBERS

GQ160460–GQ160467.

FUNDING

The Gordon and Betty Moore Foundation and the Australian Research Council (LP0668235). Funding for open access charge: internal funds from the Centre for Marine Bio-Innovation.

Conflict of interest statement. None declared.