Chloroflexi CL500-11 Populations That Predominate Deep-Lake Hypolimnion Bacterioplankton Rely on Nitrogen-Rich Dissolved Organic Matter Metabolism and C₁ Compound Oxidation

A plankton survey system (PSS) was continuously lowered and raised at ∼0.25 m s⁻¹ in a sinusoidal path from 1 to 2 m beneath the lake surface to 2 to 4 m above the bottom as the R/V Laurentian moved at ∼1.8 m s⁻¹ while logging data every 0.5 s (18). The PSS contained sensors mounted on a V fin to measure the chlorophyll a concentration (Wet Labs ECO fluorometer, Sea-Bird Scientific), the amount of photosynthetically active radiation (PAR; determined with a 4-pi sensor; model QSP2300; Biospherical Instruments Inc.), the dissolved oxygen concentration (SBE43; Sea-Bird Scientific), turbidity (Wet Labs ECO NTU; Sea-Bird Scientific), and temperature. The fluorometer output (in volts) was converted to derived chlorophyll a concentrations by regression between the fluorometer output and laboratory chlorophyll a measurements (18). Replicate samples were analyzed for the concentrations of dissolved organic carbon (DOC), particulate organic carbon (POC), particulate organic nitrogen (PON), total phosphorus (TP), particulate phosphorus (PP), and soluble reactive phosphorus (SRP) according to NOAA Great Lakes Environmental Research Laboratory (GLERL) standard operating procedures (19). Total dissolved phosphorus (TDP) values were calculated by subtracting the PP value from the TP value.

Water samples originated from 5 m below the surface and 2 to 5 m above the lake floor at the near-shore station (43°11′17″N, 86°20′38″W, where samples were obtained on 23 April, 15 July, and 23 September 2013) and the offshore station (43°11′59″N, 86°34′11″ W, where samples were obtained on 23 April, 16 July, and 24 September 2013) along the NOAA Lake Michigan-Muskegon, MI, transect (Fig. 1). Water was collected using a 30-liter Niskin bottle and prefiltered through 210- and 20-μm-mesh-size Nitex meshes into 10-liter carboys. The carboys, funnels, and meshes were cleaned with bleach, rinsed with Milli-Q water until no bleach odor remained, and rinsed twice with sample water. Prefiltered water was sequentially filtered onto 3.0-μm-pore-size polycarbonate filters and 0.22-μm-pore-size polyethersulfone filter membranes (142 mm; Millipore) using a Masterflex I/P peristaltic pump (Cole Parmer) between settings 11 and 13. Filters were folded with the bacterial biomass facing inwards and submersed into the RNAlater reagent (Ambion). Samples were stored at −20°C onboard and transferred to a −80°C freezer within 48 h of sampling. Sample filtering was limited to 10 min, and all samples were stored in RNAlater within 20 min of sampling.

Duplicate nucleic acid extractions from the same 142-mm filter membrane were performed for each of the field samples using a modified AllPrep DNA/RNA/microRNA universal kit protocol (Qiagen) (20). A part of each of the RNA fractions was converted to cDNA using a ProtoScript II first-strand cDNA synthesis kit (New England BioLabs).

DNA and cDNA were submitted to the Joint Genome Institute (JGI) for 16S rRNA gene amplicon sequencing, targeting the V4 region of the 16S rRNA gene (515F/806R universal primers) (21). Pooled libraries were sequenced on an Illumina MiSeq sequencer, using v2 chemistry and two 250-cycle (500 cycles) paired-end reads. RTA software (v1.17.28) and MCS software (v2.2.0) were used to generate data. A random subset of 40,000 read pairs was used for each of the field data samples to reduce the computational time. We used mothur software (v.1.34.3) to generate an operational taxonomic unit (OTU; 97% sequence similarity) table and rarefy data at a subsampling level that allowed inclusion of all samples (n = 4,500; certain samples contained >85% chloroplast sequences). We used the MiSeq standard operating protocol, accessed on 17 December 2014, using the SILVA (release 119) database for alignment and classification (22, 23). The OTU table was imported into Excel software, and OTU count data for the OTUs classified as members of the Anaerolineaceae were extracted for the generation of Fig. 2. All data are available on the Joint Genome Institute's genome data portal (http://genome.jgi.doe.gov/; project identifiers, 1041195 and 1041198).

The Joint Genome Institute generated metagenomic data from field samples (the 0.22- to 3-μm fraction only) collected at the offshore station in spring (at depths of 5 m [surface water; Integrated Microbial Genomes {IMG} Genomes Online Database {GOLD} accession number Ga0007769] and 108 m [deep water; IMG GOLD accession number Ga0007770]), summer (at depths of 108 m [IMG GOLD accession number Ga0007777] and 35 m [the depth with the maximum chlorophyll concentration; IMG GOLD accession number Ga0007778]), and fall (at a depth of 108 m only [low chlorophyll was present throughout the water column, so no sampling at 35 m was performed in the fall; IMG GOLD accession number Ga0007786]). Paired-end 150-bp reads were generated (HiSeq 2000 sequencing system) from ∼200-bp-fragment-size Nextera (summer and fall) or TruSeq (spring) libraries (Illumina). Metagenomic data from individual samples were assembled by use of the JGI pipeline. In brief, the BBDuk tool (filterk = 27, trimk = 27) was used to remove Illumina adapters, Illumina artifacts, and phiX and quality trim both ends to quality score Q12. Reads with quality scores averaging less than 8 over the read before trimming or with a length of under 40 bp after trimming were discarded. The remaining reads were mapped to the human HG19 reference genome with the BBMap aligner tool, with all hits with over 93% identity being discarded. Trimmed, screened, paired-end Illumina reads were assembled using SOAPdenovo software (v1.05) (24) (default settings) and multiple k-mers (81, 85, 89, 93, 97, and 101 bases). The contigs resulting from each k-mer were dereplicated using in-house Perl scripts. Contigs smaller than 1,800 bp were assembled using Newbler software (v2.8; Life Technologies) to generate larger contigs (flags were -tr, -rip, -mi 98, and -ml 80). Newbler contigs and SOAPdenovo contigs larger than 1,800 bp were combined using the minimus (v2) genome assembler (25) (flags were -D MINID = 98 and -D OVERLAP = 80). Exact duplicate and fully contained sequences were removed using the dedupe tool (github.com/datamade/dedupe).

Full 16S rRNA gene reconstruction was performed with EMIRGE software (with default settings, except for a −l setting of 150, a −i setting of 200, a −s setting of 50, and a −j setting of 1.0), using as a reference the SILVA (release 119) database clustered at 97% using the USEARCH algorithm (26). The −j setting of 1.0 allows the highest possible phylogenetic resolution of the analysis, as only sequences that are 100% identical are combined during the iterative read recruitment process. Sequences matching 18S rRNA gene sequences were excluded.

To achieve improved population genome reconstructions, we used the GAM-NGS tool (27) to sequentially merge sample-specific assemblies from deep samples at the offshore station in spring, summer, and fall (see Table S2 in the supplemental material). All resulting contigs of >3,000 bp were clustered using ESOM tools, on the basis of the tetranucleotide frequency of the contigs (28). To help extract contigs of interest (i.e., those for Chloroflexi), contigs were phylogenetically classified by searching the open reading frames identified by the Prodigal program (29) against those in the NCBI NR database using the RAPsearch2 program (30). We expanded the putative Chloroflexi ESOM bin with contigs of >1,000 bp that had a RAPsearch2 Chloroflexi classification and reduced it to the CL500-11-LM (where LM refers to Lake Michigan) sequence bin using the Maxbin method (31) and the mmgenome R library (32) (see Table S2 in the supplemental material). The latter relied on read recruitment with the Bowtie 2 program (default settings) (33) using 10 million paired-end reads from offshore spring surface and bottom water samples and summer bottom water samples. Analysis was performed with the CheckM tool to assess the completeness and purity of the final Chloroflexi CL500-11-LM sequence bin (34). We also performed a CheckM analysis on the genome of the closest sequenced isolate, Anaerolinea thermophila UNI-1 (35). The CL500-11-LM bin was submitted to the DOE Joint Genome Institute Integrated Microbial Genomes for annotation (GOLD analysis ID Ga0063436). To remove redundancy due to incomplete assembly merging, a self-BLASTP analysis of the CL500-11-LM proteins was performed to search for replicate contigs by identifying blocks of proteins (2 or more) that were duplicated at >99% sequence identity. Genes for redundant proteins were removed from the genome.

We used the STAMP program to identify genes for over- and underrepresented protein functions in CL500-11-LM relative to their representation in a combined set of 6 freshwater heterotroph genomes (listed in Table S6 in the supplemental material), A. thermophila, and the deep-water summer offshore station metagenomic data set (36). Analyses were performed using the cluster of orthologous groups of proteins (COGs) and Pfam profiles obtained from the JGI IMG data web portal. All data sets were rarefied to the lowest number of profiles (typically, the number present in CL500-11-LM). Two-sample statistical analysis to identify genes for functions that were present in CL500-11-LM at relative abundances significantly different from their abundances in the reference genomes or metagenome was performed in STAMP using Fisher's exact test for Pfam profiles (due to low numbers of occurrences of many Pfam profiles) and the chi-square test for COGs.

The Joint Genome Institute generated metatranscriptomic data from RNA extracted from the offshore station samples (the 0.22- to 3-μm fraction only) collected in spring (at a depth of 5 m during the daytime [IMG GOLD accession number Ga0007745] and nighttime [IMG GOLD accession number Ga0007744] and at a depth of 108 m during the daytime [IMG GOLD accession number Ga0007746]), summer (at a depth of 108 m during the nighttime [IMG GOLD accession number Ga0007753] and at a depth of 35 m during the daytime [IMG GOLD accession number Ga0007754]), and fall (at a depth of 108 m during the nighttime [IMG GOLD accession number Ga0007762]). rRNA was removed from 10 ng of total RNA using a Ribo-Zero rRNA removal kit (Bacteria) (Epicentre). Stranded cDNA libraries were generated using an Illumina TruSeq stranded RNA LT kit. The rRNA-depleted RNA was fragmented and reverse transcribed using random hexamers and SuperScript II reverse transcriptase (Invitrogen) followed by second-strand synthesis. The fragmented cDNA was treated with end-pair, A-tailing, adapter ligation, and 15 cycles of PCR. Sequencing (two times for 150-bp sequences) was performed on the Illumina HiSeq 2000 sequencer using a TruSeq SBS sequencing kit (v3).

Raw sequences were (i) quality trimmed to Q10 and adapter trimmed using BBDuk (with options of ktrim = r k = 25 mink = 12 tpe = t tbo = t qtrim = r trimq = 10 maq = 10 maxns = 3 minlen = 50), (ii) filtered for process artifacts using BBDuk (with an option of k = 16), and (iii) mapped against a trimmed version of the SILVA database to remove remaining rRNA reads using BBMap (options were fast = t minid = 0.90 local = t), and (iv) human reads were removed using BBMap. Remaining reads were recruited to the curated CL500-11-LM coding genes using Bowtie (v2) (with options of -local -p 10 -D 5 -R 1 -N 0 -L 25 -i S,1,2.00) (33). For samples with a high abundance of CL500-11, 10 million paired-end read pairs were used (samples collected from deep water at night in summer and fall), while all reads were used for the remaining data sets (samples collected during the day in the summer for determination of the maximum chlorophyll concentration, samples collected from the surface during the day and at night in the spring, and samples collected from deep water during the day in the spring). Mapped reads were scaled to the data set with the lowest number of recruited reads as described previously (37), and differential expression was detected using the DEseq2 package, which identifies differentially expressed genes on the basis of empirical Bayes shrinkage of both gene count variance and the fold change estimate, two problematic issues of count-based data with a large dynamic range, such as transcriptomics data (38). To account for type I error, correction of the P values using the Benjamini-Hochberg false discovery rate correction available in the DeSeq2 package was performed. The overrepresentation of functions among highly expressed genes (the functions of proteins encoded by genes with the 5%, 10%, and 25% highest levels of expression relative to the functions of proteins encoded by genes present in the same proportion in the whole CL500-11-LM genome), as well as among genes differentially expressed between surface water and deep water samples, was determined by applying a chi-square test for COG categories and a Fisher exact test for Pfam categories (using the STAMP program). The P values were corrected using the Benjamini-Hochberg false discovery rate correction.

We used the CARD-FISH probes (CL500-11-specific horseradish peroxidase-labeled probe CLGNS-584 and helper probe CLGNS-567h to augment the signal strength) and the protocol developed by Okazaki et al. (14) on subsamples consisting of the 0.22- to 20-μm fraction from the water sampled for DNA and RNA extractions. Filters were examined by fluorescence microscopy by taking a photo and counting the number of DAPI (4′,6-diamidino-2-phenylindole)-stained and probe-tagged cells within the field; a minimum of 300 DAPI-stained cells (5 to 10 fields of view) was counted per sample. Cell size and the proportion of cell pairs were determined for ∼50 cells per sample using the ZEN software measurement tools (Zeiss, Inc.). Because CL500-11 cells have a curved rod shape, the cell sizes were measured from the middle of the cell's width, following the cell shape length-wise. We identified the proportion of all cells that were part of dividing cells (cell pairs) and separately measured the length of each member of the cell pair. For all samples, cells were measured in either a DAPI- or green fluorescent protein-stained field, depending on the clarity of the cell boundaries.

All samples originated from the Muskegon transect on Lake Michigan, a long-term observatory operated by NOAA GLERL (Fig. 1). The water column was isothermal in spring and stratified in summer and fall (Fig. 1D to F), it remained oxygenated throughout the year (Fig. 1G to I), and at midday, PAR was high in surface waters and low in waters below depths of 60 to 80 m (Fig. 1M to O). N, P, and C levels were the highest in the near-shore water in spring and the lowest in the offshore deep water (Fig. 1C). The phytoplankton concentration was measured by determination of the chlorophyll a concentration and was low except near the shore in spring and offshore in summer, at the depth where chlorophyll concentrations were highest (35 m) (Fig. 1J to L).

rRNA sequencing data from the cDNA and DNA libraries were combined in one analysis, as DNA from summer offshore deep-water samples failed to be amplified for 16S rRNA iTag analyses. After rarefaction, 306 OTUs were classified as Chloroflexi, which contributed 1.8% of all rarefied sequencing reads across all samples (Chloroflexi was the 7th most abundant phylum; Proteobacteria was the most abundant phylum, accounting for ∼40% of all rarefied reads). Only one of these OTUs reached a relative abundance of >3% in any of the individual sample data sets. This OTU was classified at the family level as Anaerolineaceae and contained ∼75% of all Chloroflexi sequences identified in the entire data set. This population was primarily found in the free-living fraction of the deep offshore waters (Fig. 2) and contributed up to 20.7% of all sequences in a given sample (i.e., the 0.22- to 3-μm fraction from offshore, deep-water samples collected in the fall). To obtain more resolved taxonomic information for this OTU, we used the EMIRGE program to reconstruct small-subunit rRNA genes from the metagenomic data for the 0.22- to 3-μm fraction in samples collected in the summer at the offshore hypolimnion station. We reconstructed six unique Chloroflexi 16S rRNA genes (reconstructed from reference sequences of two members of the Anaerolineaceae, two members of the SL56 marine group of bacteria, one member of KD4-96, and one member of the genus Roseiflexus), and these 16S rRNA genes contained a total of 10.1% of all reads used to reconstruct bacterial sequences of >500 nucleotides. One sequence was generated using 7.3% of all recruited reads (see Table S1 in the supplemental material) and shared 99.6% identity to the sequence of clone CL500-11 from Crater Lake, OR, USA (GenBank accession number AF316759) and to a partial sequence retrieved from Lake Biwa, Japan (GenBank accession number AB686531).

While they were predominantly present in the 0.22- to 3-μm fraction, CL500-11-like 16S rRNA gene sequences were detected in the 3- to 20-μm fraction in fall, and CL500-11-like 16S rRNA gene sequences had nearly equal relative abundances in both fractions in the offshore surface waters in spring (Fig. 2). To obtain clues regarding this apparent shift in cell size, we performed CARD-FISH experiments with three offshore samples (the 0.22- to 20-μm fraction): surface water collected in the spring and bottom water collected in the summer and fall. The CARD-FISH data confirmed the patterns observed by 16S rRNA V4 region iTag sequencing (Fig. 2). The absolute abundance of CL500-11-like cells varied from 3.1 × 10⁴ ± 1.5 × 10⁴ cells/ml (95% confidence interval [CI]) in spring surface waters to 2.2 × 10⁴ ± 0.8 × 10⁴ and 4.6 × 10⁴ ± 1.3 × 10⁴ cells/ml in summer and fall deep-water samples, respectively. The average cell size was larger in spring surface waters (0.87 ± 0.06 μm [90% CI]) than in summer and fall surface waters (0.76 ± 0.04 μm) (see Fig. S1 in the supplemental material). In addition, the proportion of cells that appeared to be connected in cell pairs in spring was larger than that in summer and fall, though the difference was not significant (71.4% versus 50.1% of all cells; P > 0.10) (Fig. 2).

In an effort to maximize the completeness of the genome, we merged three Lake Michigan metagenomic data set assemblies and subsequently used tetranucleotide and differential coverage binning methods (see Fig. S2 to S4 in the supplemental material) to isolate contigs comprising the genomic data set for the CL500-11-LM population. Initially, we selected 503 contigs that had an average GC content of 60.9% and that carried 2,579 protein-coding genes (see Table S2 in the supplemental material). CheckM estimated the genome to be 84.5% complete with only 0.8% contamination not attributable to strain heterogeneity (see Table S3 in the supplemental material). After removal of redundancy resulting from the assembly merging process, we reduced the CL500-11-LM genome to 398 contigs carrying 2,153 protein-coding genes. CheckM indicated that this significantly reduced within-genome redundancy (see Table S3 in the supplemental material). The genome of the closest fully sequenced relative (A. thermophila UNI-1) was estimated to be 93.6% complete (CheckM). Corrected for the A. thermophila CheckM completeness estimate gap, we estimated that the CL500-11-LM bin was 90.2% complete.

The curated CL500-11-LM sequence bin contained genes for aerobic oxidative phosphorylation (see Table S4 in the supplemental material) and most genes required for glycolysis and the TCA cycle (see Table S4 in the supplemental material), including the gene for one of two key enzymes for the glyoxylate shunt (malate synthase; IMG locus tag Ga0063436_10633). Missing were genes for orthologs of the enzymes catalyzing the phosphorylation of glucose, though genes for multiple sugar kinases and the conversion of malate to oxaloacetate were present. We also identified multiple ABC transporters for sugars and peptides (see Tables S5 and S6 in the supplemental material). No evidence of a carbon fixation capability was found, but we did identify genes involved in C₁ compound oxidation: (i) type I and type II carbon monoxide dehydrogenases, on the basis of both motif and gene neighborhood analyses (see Table S7 in the supplemental material), and (ii) part of the pathway for the tetrahydrofolate-dependent oxidation of methanol, glycine, methylamines, and, potentially, betaine (four copies of trimethylamine:corrinoid methyltransferases [39]) (see Table S8 in the supplemental material). Missing genes were present in multiple copies in the metagenomic data set, though they were not part of the Chloroflexi bins. No genes encoding photosynthetic capabilities were found. However, a proteorhodopsin gene was encoded on a contig located on the margin of the Chloroflexi ESOM bin (see Fig. S3 in the supplemental material), and read mapping across multiple data sets supported its inclusion in the CL500-11-LM bin (see Fig. S4 in the supplemental material). The retinal biosynthesis pathway was only partially present, and genes encoding beta-carotene 15,15′-monooxygenase (bcmo), and lycopene cyclase (crtY) were missing from the CL500-11-LM bin on the basis of analysis of the gene annotation and a BLASTP search with the corresponding protein sequences from “Candidatus Pelagibacter ubique” HTCC1062 (E value, <1e−5).

Sixty-five percent of CL500-11-LM genes encoding proteins did not have an ortholog in the most closely related sequenced Chloroflexi isolate, A. thermophila UNI-1 (an ortholog was defined as the reciprocal best hit with a BLASTP score ratio of >0.3, which corresponds to 30% identity across 70% of the length of the protein [40]). The genome of CL500-11-LM encoded significantly fewer genes for regulatory processes and carbohydrate metabolism than the genome of A. thermophila, while genes for amino acid and coenzyme transport and metabolism and cell motility were overrepresented in CL500-11-LM relative to the representation of genes with the same functions in A. thermophila (Fig. 3; see also Table S9 in the supplemental material). The latter was the result of the unique presence in CL500-11-LM of genes for the flagellar apparatus and two methyl-accepting chemotaxis proteins. In line with limited regulatory mechanisms in CL500-11-LM, only five RNA polymerase sigma factors were identified (a sigma 28 factor, a sigma 54 factor, and 3 sigma 70 factors), whereas 12 sigma factors and 2 anti-sigma factors were identified in the genome of A. thermophila UNI-1.

Genes involved in cell motility, amino acid transport and metabolism, and cell wall and membrane biosynthesis were overrepresented in CL500-11-LM compared to their representation in the genomes of a set of six other abundant freshwater heterotrophic bacteria with relatively streamlined genomes (<2.5 Mbp; two acI-lineage genomes, two LD12-lineage genomes, and two Polynucleobacter [PnecC bacterium] genomes; see Tables S6 and S9 in the supplemental material) and the metagenome for deep water from the offshore station in summer (Fig. 3). A more detailed comparison using the Pfam profiles indicated a variety of peptide and carbohydrate transporters, peptidases, methyltransferases, and glycosyltransferases to be overrepresented in CL500-11-LM relative to their representation in its closest sequenced relative and the community that it dominates (Fig. 3; see also Table S9 in the supplemental material). The DOM substrate uptake of CL500-11-LM most closely resembled that of acI (see Table S6 in the supplemental material), though it was more diverse and marked by much higher levels of redundancy of di- and oligopeptide transporters (31 proteins for CL500-11-LM versus 5 for acI). The presence of multiple copies of CO dehydrogenase genes was another unique feature of CL500-11-LM.

We recruited metatranscriptomic reads to the CL500-11-LM coding sequences for all data sets where iTag analysis indicated the presence of CL500-11-LM (see Tables S5 and S10 in the supplemental material). Average expression levels in samples from offshore stations (surface and deep waters in the spring, water from the depth with the maximum chlorophyll concentration and deep water in the summer, and deep water in the fall) indicated that among the top 5% most highly expressed genes, the following genes were overrepresented relative to their prevalence in the whole CL500-11-LM genome: genes for amino acid transport and metabolism (14.5% of all highly expressed genes relative to 8.3% of all CL500-11-LM genes; q = 0.15), particularly genes for di- and oligopeptide transporters (5.2% versus 0.5%; q = 0.001); genes for energy production (14.5% versus 4.5%; q = 2.4 × 10⁻⁵), which included genes for both the type I and the type II carbon monoxide dehydrogenases; and genes for protein turnover (10.0% versus 2.0%; q = 2.2 × 10⁻⁶) (see Table S11 in the supplemental material). Relatively few housekeeping genes were among the most highly expressed genes, though when the top 10% and top 25% most highly expressed genes were included, genes with housekeeping functions were more highly represented, as were genes involved in motility (see Table S11 in the supplemental material). Genes with transport functions for DOM were overrepresented in the CL500-11-LM transcriptome relative to the prevalence of genes with these functions in the CL500-11-LM genome (11.4% of all mRNA reads; see Table S6 in the supplemental material).

The highest number of differentially expressed genes was identified when samples from deep water were compared to samples from surface water, regardless of the season (75 genes) (see Table S12 in the supplemental material). Only 67 genes were differentially expressed when samples from deep water in the spring were excluded; 60 of these were also identified to be differentially expressed when samples collected from deep water in the spring were included. When samples collected in spring were compared to samples from summer and fall, regardless of depth (samples collected from the surface [day and night] and deep water in the spring relative to samples collected from the depth with the maximum chlorophyll concentration and from deep water collected in the summer and fall), were compared, the result was only 7 differentially expressed genes. Finally, when samples collected from surface and deep water in the spring were compared to samples collected from deep water in the summer and fall, 26 genes were found to be differentially expressed, and 19 of these were shared by the samples collected from surface water in the spring and samples collected from deep water in the summer and fall. This indicated that the location in the water column was the factor that most influenced CL500-11-LM gene expression. Among measured environmental parameters, photosynthetically active radiation was the parameter that varied most between the surface and deep water (Fig. 1). More genes were more highly expressed in surface water than deep water, including genes for defense mechanisms, Fe-S assembly proteins (suf operon), Fe uptake, components of the electron chain, glycolysis and the tricarboxylic acid (TCA) cycle, proteorhodopsin, and proteins involved in carotenoid biosynthesis (see Table S12 in the supplemental material). Genes more highly expressed in deep water included those for multiple amino acid metabolic functions, including tryptophan biosynthesis, type III glutamine synthetase, part of a glycine cleavage complex, and part of the type II carbon monoxide dehydrogenase operon.