Genome Evolution and Nitrogen Fixation in Bacterial Ectosymbionts of a Protist Inhabiting Wood-Feeding Cockroaches

ABSTRACT

By combining genomics and isotope imaging analysis using high-resolution secondary ion mass spectrometry (NanoSIMS), we examined the function and evolution of Bacteroidales ectosymbionts of the protist Barbulanympha from the hindguts of the wood-eating cockroach Cryptocercus punctulatus. In particular, we investigated the structure of ectosymbiont genomes, which, in contrast to those of endosymbionts, has been little studied to date, and tested the hypothesis that these ectosymbionts fix nitrogen. Unlike with most obligate endosymbionts, genome reduction has not played a major role in the evolution of the Barbulanympha ectosymbionts. Instead, interaction with the external environment has remained important for this symbiont as genes for synthesis of transporters, outer membrane proteins, lipopolysaccharides, and lipoproteins have been retained. The ectosymbiont genome carried two complete operons for nitrogen fixation, a urea transporter, and a urease, indicating the availability of nitrogen as a driving force behind the symbiosis. NanoSIMS analysis of C. punctulatus hindgut symbionts exposed in vivo to ¹⁵N₂ supports the hypothesis that Barbulanympha ectosymbionts are capable of nitrogen fixation. This genomic and in vivo functional investigation of protist ectosymbionts highlights the diversity of evolutionary forces and trajectories that shape symbiotic interactions.

IMPORTANCE The ecological and evolutionary importance of symbioses is increasingly clear, but the overall diversity of symbiotic interactions remains poorly explored. In this study, we investigated the evolution and nitrogen fixation capabilities of ectosymbionts attached to the protist Barbulanympha from the hindgut of the wood-eating cockroach Cryptocercus punctulatus. In addressing genome evolution of protist ectosymbionts, our data suggest that the ecological pressures influencing the evolution of extracellular symbionts clearly differ from intracellular symbionts and organelles. Using NanoSIMS analysis, we also obtained direct imaging evidence of a specific hindgut microbe playing a role in nitrogen fixation. These results demonstrate the power of combining NanoSIMS and genomics tools for investigating the biology of uncultivable microbes. This investigation paves the way for a more precise understanding of microbial interactions in the hindguts of wood-eating insects and further exploration of the diversity and ecological significance of symbiosis between microbes.

INTRODUCTION

Bacteria form mutualistic symbioses with a great diversity of eukaryotic organisms (1–5). By providing beneficial and often essential functions for their hosts, thus expanding the range of ecological niches that can be occupied, the ecological and evolutionary importance of symbioses is increasingly clear, but the overall diversity of symbiotic interactions remains poorly explored (6–9). In particular, extracellular symbioses have been relatively understudied. These encompass incredibly diverse associations ranging from obligate to facultative for the host and/or symbiont, and include symbionts that are extracellular but inside a multicellular host, such as animal gut bacteria, as well as symbionts that reside on the surface of host cells, referred to here as ectosymbionts, such as those attached to protists (see, for example, references 10 to 14).

Genomic analyses in conjunction with culture-independent methods have provided the best insight into the functioning and evolution of bacterial symbioses. These investigations have generally focused on intracellular symbionts (referred to here as endosymbionts) of animals and, more recently, of protists, including ciliates, amoebae, and parabasalids (15–18), revealing common patterns of genome evolution. The genomes of vertically transmitted, obligate endosymbionts are typically reduced in gene content, are AT biased, and carry an elevated number of pseudogenes compared to their free-living relatives (15, 16, 19–21). These genomic features are thought to be the result of population bottlenecks, genetic drift, and relaxed selection on genes that are no longer needed as a free-living bacterium adapts to a symbiotic lifestyle (20, 22, 23).

While the endosymbionts of diverse hosts appear to share convergent evolutionary outcomes due to common selective pressures, far less is known about ectosymbionts. These associations are also common in certain environments, appear to be every bit as specific as intracellular associations, and can involve elaborate structural adaptations in both the host and symbiont (10, 24–26). The specificity and integration of these ectosymbionts suggest that these bacteria have evolved similarly to obligate intracellular endosymbionts. However, genome-wide analyses of ectosymbionts, and evidence for their functional roles, have lagged behind those of endosymbionts.

The hindguts of wood-eating lower termites and related Cryptocercus cockroaches provide an excellent model system to investigate differences between endo- and ectosymbionts and the biology of symbiosis in general. Here, a highly endemic and diverse community of microbial symbionts, including protists (mainly parabasalids and oxymonads), bacteria, and archaea, thrives solely on ingested wood lignocellulose, a highly recalcitrant and nutrient-poor substance. The symbionts produce food in the form of acetate through the fermentation of lignocellulose and essential nitrogenous nutrients for the insect host (27, 28). Within this community of hindgut symbionts, perhaps even more tightly integrated are the numerous endo- and ectosymbiotic associations between hindgut protists and bacteria. These endosymbionts (bacteria inside a protist cell) and ectosymbionts (bacteria on the surface of a protist cell) form highly specific associations often resulting in coevolution among protists, their bacterial symbionts, and also the insect hosts (29–31). In many cases, endo- or ectosymbionts of protists and free-living bacteria in the hindguts are closely related, providing the opportunity to directly compare and trace the evolution of these two modes of symbiosis (32).

Nitrogen fixation is likely an essential function for lignocellulose-digesting gut communities because of the low levels of nitrogen available from this diet. Molecular and genomic data indicate that hindgut symbionts from wood-eating termites and cockroaches possess a diversity of nitrogenase genes (16, 33–37), but actual nitrogen fixation by specific symbionts has rarely been demonstrated. For instance, a few cultivated bacteria have been shown to fix nitrogen, but these are likely not the dominant nitrogen fixers in the hindgut (37–40). As well, in using cultivation-independent methods, results of PCR, acetylene reduction, and ¹⁵N₂ incorporation assays suggest that the majority of nitrogen fixation in the hindguts of Hodotermopsis sjoestedti is performed by spirochete endosymbionts of Eucomonympha (41).

For identifying nitrogen-fixing symbionts in situ, stable isotope probing (SIP) with ¹⁵N₂ followed by high lateral-resolution imaging secondary ion mass spectrometry (SIMS) with a Cameca NanoSIMS has been shown to be a successful cultivation-independent approach (42). This combined SIP-NanoSIMS approach (also known as nanoSIP) allows direct imaging of individual cells based on their isotopic composition, thereby indicating which cells have used an isotopically labeled substrate (e.g., ¹⁵N₂). A similar approach was also successful for examining carbon transfer between protists and their symbionts in termite hindgut communities using ¹³C-labeled cellulose (43).

In this study, we combined genomics and NanoSIMS analysis to examine the function and evolution of Bacteroidales ectosymbionts of Barbulanympha, a lignocellulose-digesting parabasalid protist from the hindgut of the wood-eating cockroach Cryptocercus punctulatus. Bacteroidales have a predilection for forming intracellular and extracellular associations, and complete genomes are available from endosymbiotic, host-associated, and free-living relatives, together providing a unique opportunity to investigate the evolution of diverse lifestyles within a single bacterial order. We examined the beneficial functions that Barbulanympha ectosymbionts may provide—in particular, whether they fix nitrogen. We also examined the selection pressures faced by ectosymbionts compared to endosymbionts of protists and the evolutionary consequences of an ectosymbiotic lifestyle.

MATERIALS AND METHODS

Cockroach collections and identification of Barbulanympha from the hindgut.

Cryptocercus punctulatus adults and nymphs were collected from Mountain Lake Biological Station, Giles County, VA (37.364°N, 80.519°W). Adults collected in September 2011 and September 2012 were used to isolate single Barbulanympha cells for genomic sequencing. Nymphs collected in April 2011 were used for stable isotope labeling experiments, scanning electron microscopy (SEM), and NanoSIMS.

Using light microscopy (for genomics) or SEM (for NanoSIMS), Barbulanympha cells were visually identified from the C. punctulatus hindgut based on published descriptions (13, 24, 44). Barbulanympha has a distinct morphology compared to other large protists in the hindgut (e.g., Trichonympha and Eucomonympha).

DNA isolation and library preparation from single Barbulanympha cells.

To isolate cells for sequencing, the hindgut was dissected from live cockroaches and the contents were collected into Trager's solution U (45). Viewed under an inverted light microscope, Barbulanympha cells were individually transferred to clean Trager's solution 2 times to remove unattached bacteria and then into a microcentrifuge tube using pulled glass capillary tubes attached to a syringe via rubber tubing. Three cells (named Barb4, Barb6, and Barb7) and one cell (Barb6XT) were isolated from a cockroach collected in September 2011 and September 2012, respectively. From each cell, approximately 1 ng of DNA was isolated using the MasterPure DNA extraction kit (EpiCentre, Madison, WI).

From the DNA of one cell (Barb6XT), a sequencing library was generated using the Nextera XT kit (Illumina, San Diego, CA) and 250-bp paired-end reads were sequenced using Illumina MiSeq. Using Sickle (https://github.com/najoshi/sickle) (46), reads were trimmed where the quality score fell below 20, and only reads at least 150 bp in length were retained. Transposon sequences used to generate the Nextera XT library were also trimmed from the ends of the reads using a custom Perl script (removeTP.pl), available through https://github.com/JFP-Laboratory/Genomics.

Genomic DNA was isolated from three other cells (Barb4, Barb6, and Barb7) and amplified using phi29 DNA polymerase from the Genomiphi v2 kit per the manufacturer's protocol (GE Healthcare Life Sciences, Mississauga, ON, Canada). The DNA was mechanically sheared with an S220 focused ultrasonicator (Covaris, Woburn, MA), and sequencing libraries were constructed from 400- to 600-bp fragments using a TruSeq kit (Illumina). These libraries were sequenced using Illumina HiSeq, generating 100-bp paired-end reads. Reads were trimmed where the quality score fell below 20, and only reads at least 50 bp in length were retained.

Sequence assembly and binning.

Sequences were assembled using Ray v2.1.1-devel with kmers ranging from 25 to 95 (47), ABySS v1.5.1 with kmers ranging from 21 to 95 (48), MIRA v4.0 (49), and the CLC Genomics Workbench (Qiagen Bioinformatics, Redwood City, CA). For the Ray and ABySS assemblies, the kmer that generated the assembly with the longest contig and the longest N50 was chosen for further analysis. The assemblies generated by the different methods were compared using QUAST (50). Depth of read coverage was determined by mapping the reads to the assembled contigs using Bowtie2 (51). Based on the lengths of the contigs generated, the different methods were equivocal (see Table S1 in the supplemental material). Ray assemblies were used for all further analyses, as Ray was able to efficiently assemble the larger HiSeq data sets from Barb4, Barb6, and Barb7.

rRNA genes were predicted from the contigs by the HMMER 3.0 program (http://weizhong-lab.ucsd.edu/metagenomic-analysis/server/hmm_rRNA/) (52). rRNA sequences were also assembled using EMIRGE to assess the level of potential contamination from bacteria other than the Barbulanympha ectosymbiont (53).

Contigs greater than 1,000 bp were binned, using VizBin, based on pentanucleotide frequencies to identify contigs belonging to the ectosymbiont, distinguishing them from host-derived contigs, other bacteria, and contaminants (54). From the VizBin plot, two bins were identified that included most of the assembled contigs (see Fig. S1 in the supplemental material). All contigs containing 16S rRNAs were contained within one of the bins, and these 16S rRNA sequences were at least 99% similar to previously sequenced Bacteroidales ectosymbionts of Barbulanympha, so these contigs were designated the ectosymbiont contigs. The contigs within this bin were analyzed again using VizBin, but the plot did not reveal clusters to further bin these contigs. These contigs were curated by using BLASTn to identify contaminating contigs. Contigs with best BLASTn hits (E value < 10⁻⁵) to plasmid vectors, eukaryotes (potentially host derived), or Escherichia coli in GenBank's nucleotide database were excluded.

The second bin contained contigs containing 18S rRNA genes matching Barbulanympha, so these contigs were designated host contigs. The GC content of all contigs was analyzed to provide further evidence distinguishing the taxonomic origin of the ectosymbiont and host contigs.

Genome annotation and genome completeness.

Open reading frames (ORFs) were identified and annotated using Prokka v1.5.2 (55). The KEGG Automatic Annotation Server (http://www.genome.jp/kegg/kaas/) (56) and reverse position-specific (RPS) BLAST searches of the Cluster of Orthologous Groups (COG) database (http://weizhong-lab.ucsd.edu/metagenomic-analysis/server/cog/) were also used to obtain functional annotations.

Genome completeness was estimated for each set of ectosymbiont contigs using a set of 139 conserved single copy genes (CSCG) determined from all finished bacterial genomes (57). Hidden Markov model profiles (HMMs) of the protein families (Pfam) of these genes were used to search the ORFs from each set of ectosymbiont contigs using HMMER3 (58). To calculate genome completeness, the number of CSCG Pfams found in the ectosymbiont contigs that scored above precalculated cutoffs was determined and then divided by the total number of CSCG Pfams (57). The number of significant hits to each CSCG Pfam indicated the copy number of these genes in each set of ectosymbiont contigs.

Genome comparisons.

The ORFs annotated from each set of ectosymbiont contigs were compared by clustering the ORFs into homologous gene families using the COGtriangles algorithm (59). ORF clusters unique to an ectosymbiont were further analyzed by BLASTp searches (E value < 10⁻⁵) to a database of all Barbulanympha ectosymbiont ORFs and to the GenBank nonredundant (nr) nucleotide database.

Polymorphism analysis.

Sequence reads from the 4 data sets were mapped to the Barb6XT contigs using Bowtie2 (51), including mapping the Barb6XT reads to the Barb6XT contigs to examine polymorphisms among the ectosymbionts on a single host cell. The indels were realigned to the contigs using Genome Analysis Tool Kit (GATK) v3.1-1 (60), and single nucleotide polymorphisms (SNPs) were identified using samtools v0.1.18 (61) and snpEffect v4.0b (62). Polymorphisms were identified relative to the Barb6XT ectosymbiont contigs.

Phylogenetic analyses.

16S rRNA sequences annotated from the ectosymbiont contigs were aligned using MAFFT L-INS-i (63) with 16S rRNA sequences obtained from GenBank representing the diversity of Bacteroidales from clusters most closely related to the Barbulanympha ectosymbionts. A separate alignment was made for 18S rRNA sequences identified from the nonectosymbiont contigs. From each alignment, short sequences (<400 bp) were removed, as were variable and ambiguously aligned sites using Gblocks (http://molevol.cmima.csic.es/castresana/Gblocks_server.html) (64). Phylogenetic trees were generated from a maximum likelihood analysis using RAxML v8.1.3, implementing a general time-reversible (GTR) model of nucleotide substitution with the Γ (gamma) model of rate heterogeneity (65). Statistical support for the best maximum likelihood tree was assessed from 1,000 bootstrap replicates.

To generate a NifH phylogeny, the amino acids of ORFs annotated as “nitrogenase iron protein” or nifH were aligned using MAFFT with amino acid sequences from GenBank representing the known diversity of NifH, including those from recently sequenced Bacteroidales genomes (66). The alignment was trimmed using GBlocks. Using RAxML, LG was determined to be the best protein model and a maximum likelihood phylogenetic tree was calculated with 1,000 bootstrap replicates. Phylogenies were also obtained for the amino acid sequences of NifD-like nitrogenase subunit, NifK-like nitrogenase subunit, urease subunit A, urease subunit B, and the substrate binding proteins of an ABC-type transporter.

PCR analysis of potential genome rearrangements.

BLASTn searches (E value < 10⁻¹⁵) were used to locate syntenic regions and potential rearrangements among the ectosymbiont contigs from different host cells. To exclude the possibility that these rearrangements were due to assembly errors, two potential rearrangement sites were confirmed by PCR and sequencing. For each of the two sites, PCR primers were designed to anneal on either side of the rearrangement sites and used to amplify DNA extracted from the hindgut contents of the same Cryptocercus punctulatus cockroaches as used to isolate the Barbulanympha cells (MasterPure DNA extraction kit; EpiCentre, Madison, WI).

For the first site, primers 114F (5′ TCT TGT CGG GGA TGG TAG TAG) and 116R (5′ AAT GGG CTT GGA TTT CGA TGA G), specific for the Barb4, Barb6, and Barb7 ectosymbiont genomes, were designed to anneal to an ABC-type transporter permease and a transposase ORF, repectively. Primer 114F was also paired with primer 1461R (5′ AAA GGG CGC GAT TGG TAT G), designed to anneal to a hypothetical ORF and amplify the Barb6XT ectosymbiont genome. A second site was analyzed using primers 172F (5′ CTA TGC ACA TAT TCG CGA CAT C) and 174R (5′TCC AGG AGA AGA GAC GAA AC), which anneal to siroheme synthase and an intergenic region, respectively, and were designed to amplify the Barb4 and Barb7 ectosymbiont genomes. Primer 172F was also paired with primer 683R (5′ TCC AGG AGA AGA GAC GAA AC), designed to anneal to a transposase and amplify the Barb6 ectosymbiont genome. Standard thermal cycling conditions were used (94°C for 2 min, followed by 30 cycles of 94°C for 30 s, 56°C for 1 min (site 1 primers) or 53°C for 45 s (site 2 primers), and 72°C for 1 min or 45 s, ending with a final extension at 72°C for 10 min). The resulting PCR products were Sanger sequenced (Thermo Fisher Scientific, Waltham, MA).

NanoSIMS analysis of Barbulanympha exposed to ¹⁵N₂.

Three live C. punctulatus cockroach nymphs were placed in 150-ml glass Wheaton vials with moist cellulose. The air in the vials was equilibrated with ¹⁵N₂ gas (Cambridge Isotope Laboratories, Tewksbury, MA) to give a final ¹⁵N₂/¹⁴N₂ ratio of 0.38 and an O₂ content of 26%. During the incubation, the air was not refreshed and the gas lines were closed to the external atmosphere. Following a 2-week exposure to ¹⁵N-enriched air, the cockroaches were removed from the vials and dissected, and the hindgut material was pooled.

As previously described for whole-cell SEM and NanoSIMS imaging, the hindgut material was fixed with 2% glutaraldehyde, postfixed in 1% OsO₄, and then gently filtered onto 1-μm-pore membrane filters. The filters were treated with a series of ethanol dehydration steps (50, 70, 90, and 100% ethanol), dried in a carbon dioxide critical point drier, affixed to an SEM stub, and sputter coated with 5-nm iridium (43). SEM was used to identify Barbulanympha cells, free-living bacteria, and other protists from the hindgut material to target for NanoSIMS analysis. The locations of these targets were mapped on the SEM stubs using a JEOL 7401F or an FEI Inspect F scanning electron microscope.

High lateral-resolution imaging secondary ion mass spectrometry (SIMS) was performed with a Cameca NanoSIMS 50 (Gennevilliers, France) to directly image the nitrogen isotopic composition of individual cells. High relative ¹⁵N enrichment indicates fixation of ¹⁵N₂ or direct access to newly fixed ¹⁵N. NanoSIMS was performed with a focused 10-pA, 16-keV Cs⁺ primary ion beam scanned over the sample to eject secondary ions to generate quantitative isotopic images. Scans were between 10 by 10 and 20 by 20 μm² with 256 by 256 pixels and a dwell time of 1 ms/pixel. Secondary ions were collected for ¹⁵N¹²C⁻ and ¹⁴N¹²C⁻ using electron multipliers in pulse counting mode (43). Secondary electrons were collected for sample visualization. Targets were first sputtered to a depth of >50 nm before data collection to enhance ion yield. Twenty to 30 serial scans were made for each imaged area to obtain sufficient ion counts for the minor isotope.

NanoSIMS ion image data were processed with a custom software package (LIMAGE; L. R. Nittler, Carnegie Institution of Washington, Washington, DC) run with PV-Wave (Rogue Software, Boulder, CO). Quantitative ¹⁵N/¹⁴N isotope ratio images were generated from the ¹⁵N¹²C⁻ and ¹⁴N¹²C⁻ ion images. Isotope data for Barbulanympha, its ectosymbionts, and other protists were extracted from hand-drawn regions of interest (ROIs) based on secondary ion and electron images, excluding areas of low ion counts (<5% of maximum). Isotope data for areas of free-living bacteria were extracted based on ¹⁴N¹²C⁻ images using the LIMAGE automated particle-finding software (verified manually). The data are expressed as ¹⁵N/¹⁴N isotope ratios and atom percent excess (APE) ¹⁵N (APE = [R_f/(R_f + 1) − R_i/(R_i + 1)] × 100, where R_f is the measured ratio and R_i is the initial ratio) (67). The R_i was determined to be 0.00367 based on uncorrected NanoSIMS measurements. ROIs are determined to be significantly enriched if the R_f is more than 3 standard errors greater than the R_i. The Student t test was used to assess the enrichment differences between Barbulanympha cell interiors and their ectosymbionts on a pairwise basis to allow for natural variation in activity among these consortia.

Accession number(s).

The ectosymbiont contigs from each assembly were deposited at DDBJ/EMBL/GenBank under BioProject accession number PRJNA278755, BioSample accession numbers SAMN03428871, SAMN03428873, SAMN03428874, and SAMN03428875, and GenBank accession numbers LBCW00000000, LBCX00000000, LBCY00000000, and LBCZ00000000. The versions described in this paper are versions LBCW01000000, LBCX01000000, LBCY01000000, and LBCZ01000000.

RESULTS

The DNA sequences obtained from single Barbulanympha cells were assembled (see Table S1 in the supplemental material) and a single full-length (or nearly so) 16S rRNA sequence was predicted from each assembly. Truncated 16S rRNA sequences were also predicted from the ends of two Barb6XT and one Barb7 contig, and these exactly matched the longer 16S rRNA sequences or did not overlap. All of the 16S rRNA sequences shared at least 99% identity to previously sequenced 16S rRNA clones from Bacteroidales ectosymbionts of Barbulanympha (Fig. 1). Fluorescent in situ hybridization (FISH) has been previously used to confirm that these sequences, representing the dominant 16S rRNA phylotype from isolated Barbulanympha cells, belong to the ectosymbionts of Barbulanympha (81), Based on EMIRGE reconstructions and HMM searches, no other 16S rRNA genes were identified, again indicating that Bacteroidales ectosymbionts were the dominant bacteria associated with Barbulanympha and were the only bacteria sequenced to a significant depth.

FIG 1.

Comparison of the phylogenies of Bacteroidales ectosymbionts and their Barbulanympha hosts. Maximum likelihood trees of 16S rRNA sequences of the ectosymbionts from single Barbulanympha cells (on the left) are matched with the 18S rRNA sequences of their host cell (on the right). Light microscopic images of the four isolated cells are shown between the trees, and gray lines link to the sequences in the trees obtained from each cell. Bootstrap support is shown for branches with greater than 70% support. New sequences are in bold, as are Bacteroidales that are known to have nitrogen fixation genes. Scale bar = 50 μm.

The identity of the host cells as Barbulanympha was confirmed from 18S rRNA sequences predicted from the assembled contigs. The Barbulanympha 18S rRNA sequences from Barb4, Barb6, and Barb6XT shared at least 99.3% identity with each other, while that of Barb7 was more divergent, sharing a mean of 95.1% identity with the other three. Phylogenetic analysis confirmed that the Barb7 host groups with a different clade of Barbulanympha (Fig. 1).

The assembled contigs were binned to identify a set of contigs belonging to the ectosymbiont. A distinct bin of contigs was identified that contained all annotated 16S rRNA genes, so these contigs were designated as belonging to the ectosymbiont (see Fig. S1 and Table S2 in the supplemental material). The contigs within this bin also had a distinct GC content, clearly distinguishing them from the remaining contigs, most of which likely belong to the host (see Fig. S2 in the supplemental material). The ectosymbiont contigs were further curated to exclude six contigs hitting Escherichia coli with nearly 100% similarity and one contig that matched a plasmid vector from the Barb6 ectosymbiont contigs. Seven contigs that had a best BLASTn hit to a eukaryote were also excluded (2 from Barb4, 3 from Barb6, and 2 from Barb7).

Completeness of the ectosymbiont genome.

As expected for the genome of one organism, known single-copy genes were found only once in a set of ectosymbiont contigs with the exception of 2 hits to isopentanylpyrophosphate (IPP) transferases (PFAM01715) in Barb6XT, Barb4, and Barb6. Two versions of this gene are also found in the genomes of closely related Bacteroidetes (e.g., Parabacteroides distasonis, Parabacteroides merdae, and Tannerella forsythia). This result indicates that each set of ectosymbiont contigs was from a single bacterial genome and not a mixture of genomes from multiple organisms.

The best assembly resulted from Barb6XT. Genome completeness of the Barb6XT ectosymbiont contigs is estimated at 94.2%; therefore, these contigs contain nearly the complete coding capacity of the bacterial ectosymbiont (Table 1). The total length of the Barb6XT ectosymbiont contigs, the GC content, and the number and density of ORFs fell within the range of free-living and host-associated relatives (Table 1). The Barb4 ectosymbiont contigs also represented a nearly complete genome (92.1%) with a similar total length and GC content (see Table S2 in the supplemental material). The Barb6 and Barb7 ectosymbiont contigs were less complete (87.8 and 61.9%, respectively) but were similar in GC content.

TABLE 1.

Comparison of Barbulanympha ectosymbiont Barb6XT genome features with related Bacteroidales

Bacterium	Reference	GenBank accession no.	Lifestyle	Habitat	Genome size (Mb)	GC content (%)	# CDS	Coding density (%)	Genome completeness (%)a
Barb6XT	This study	LBCW00000000	Attached extracellularly, mutualist	Barbulanympha in cockroach hindgut	3.43	48.7	3,133	82.3	94.2
Parabacteroides distasonis ATCC 8503	11	NC_009615	Host associated, mutualist	Human gut	4.81	45.1	3,849	90.1	98.6*
Tannerella forsythia ATCC 43037	Unpublished data	NC_016610	Host associated, pathogenic	Human periodontal pocket	3.41	47	2,665	84.7	98.6*
Paludibacter propionicigenes WB4	68	NC_014734	Free-living	Plant residue from rice field soil	3.69	38.9	2,967	84.4	99.3*
“Candidatus Azobacteroides pseudotrichonymphae” CFP2	16	NC_011565	Intracellular, mutualist	Pseudotrichonympha in termite hindgut	1.11	33	758	71.2	95.0*

^aThe presence of a set of conserved single copy genes was used to estimate genome completeness in Barb6XT and for comparison also estimated in complete genomes (*).

The ectosymbiont genome content more closely resembles that of free-living bacteria than an endosymbiont from the same environment.

The genes annotated from the Barbulanympha ectosymbiont genomes were compared with those of other Bacteroidales to detect functionally important differences due to different lifestyles (Table 1; Fig. 1) (11, 16, 68). The Barb6XT ectosymbiont encoded many of the core functions expected from both free-living and symbiotic bacteria, including a complete set of ribosomal proteins except for L17 (L17 was found on a “nonectosymbiont” contig), DNA replication genes, and tRNA synthetases. Genes to synthesize all amino acids were also present, except for the chorismate mutase gene, which is needed to synthesize tyrosine and phenylalanine (Fig. 2).

FIG 2.

Metabolic pathways predicted from the Barb6XT ectosymbiont. Dotted arrows indicate pathways where an imported or synthesized product is likely to be used. For some functions, Barb6XT carries multiple COGs with similar functions (indicated by a multiplication sign). OMP, outer membrane protein; OMRP, outer membrane receptor protein; FMN, flavin mononucleotide; FAD, flavin adenine dinucleotide; CoA, coenzyme A. The pathways are colored to indicate whether they are found in other Bacteroidales that are free-living (FL; Paludibacter propionicigenes WB4), free-living but host-associated (HA; Parabacteroides distasonis ATCC 8503 and/or Tannerella forsythia ATCC 43037), or endosymbiotic (endo; “Candidatus Azobacteroides pseudotrichonymphae” CFP2), in addition to Barb6XT (ecto). For descriptions of these bacteria and their genomes, see Table 1.

Energy metabolism was also as expected for an anaerobic microbe. Genes for oxidative phosphorylation and the tricarboxylic acid (TCA) cycle were lacking, suggesting that energy is obtained mainly through glycolysis (although pyruvate kinase was absent) and potentially fumarate respiration (Fig. 2). The ectosymbiont genome did not carry genes for NADH dehydrogenase but contained an operon for five subunits of Na⁺-translocating NADH-quinone reductase (Na⁺-NQR), with a sixth subunit found on a separate contig (Fig. 2). This complex catalyzes the transfer of electrons from NADH to ubiquinone and pumps sodium ions (Na⁺) rather than protons (H⁺) (69). Six subunits encoding the related RNF complex, typically an NADH:ferrodoxin dehydrogenase, were also found in an operon (Fig. 2).

The Barbulanympha ectosymbiont genomes contained numerous ORFs shared with free-living species but absent in the endosymbiont “Candidatus Azobacteroides pseudotrichonymphae” of the protist Pseudotrichonympha grassi from the hindgut of the termite Coptotermes formosanus. Most significantly, the ectosymbionts retained genes to synthesize an outer membrane typical of free-living Gram-negative bacteria. The Barb6XT ectosymbiont genome encoded numerous efflux transporters (16 ORFs), outer membrane proteins (15 ORFs), and outer membrane receptor proteins (OMRPs; 58 ORFs), as well as retaining the ability to synthesize lipopolysaccharides and lipoproteins for the outer membrane, all of which except one efflux transporter were absent from “Ca. Azobacteroides pseudotrichonymphae” (Fig. 2). Many of the OMRPs belonged to the SusC/RagA family and were found paired with ORFs of the SusD/RagB family. These proteins are known to bind and import large oligosaccharides (70). Also like in its free-living relatives, multiple sugar transporters were encoded, as well as riboflavin synthesis (Fig. 2). More generally, Barb6XT did not exhibit reductions in inorganic ion transport and metabolism (including outer membrane proteins), cell motility, and defense mechanism genes (particularly efflux proteins) as seen in “Ca. Azobacteroides pseudotrichonymphae” (Fig. 3).

FIG 3.

Comparison of COG functional profiles for genomes of Bacteroidales related to Barbulanympha ectosymbionts (Barb6XT and Barb4). The percentage of ORFs in a COG class is relative to the total number of ORFs classifiable into a COG. Apseu, “Candidatus Azobacteroides pseudotrichonymphae” CFP2; Pdist, Parabacteroides distasonis ATCC 8503; Tfors, Tannerella forsythia ATCC 43037; Pprop, Paludibacter propionicigenes WB4.

Barb6XT was reduced in the relative number of COGs relating to transcription, cell wall biosynthesis, and carbohydrate transport and metabolism similarly to “Ca. Azobacteroides pseudotrichonymphae” compared to free-living Bacteroidales. This genome was also reduced in signal transduction mechanisms but not to the same extent as “Ca. Azobacteroides pseudotrichonymphae” (Fig. 3).

Interestingly, an increase in COGs related to replication, recombination, and repair was observed, primarily due to an increase in the number of transposases (64 ORFs assigned to 6 transposase COGs plus 54 unassigned ORFs annotated as transposases) (Fig. 3; see also Table S4 in the supplemental material). The free-living Bacteroidales encoded less than half of the number of transposases, and these were assigned to fewer and to different COG categories (see Table S4 in the supplemental material). Only 1 out of 6 transposase COGs in Barb6XT (COG3547) was also found in the other Bacteroidales. No transposases were found in “Ca. Azobacteroides pseudotrichonymphae.”

Nitrogen recycling and fixation are key symbiont functions.

The genomes of Barbulanympha ectosymbionts and the endosymbiont “Ca. Azobacteroides. pseudotrichonymphae,” both symbionts of parabasalid protists in the hindguts of related wood-eating insects, possess genes involved with nitrogen recycling (i.e., to recover fixed nitrogen, usually ammonia, from waste products such as urea or uric acid). These symbionts are the only Bacteroidales known to encode subunits for an ABC-type transporter for urea, as well as a urease that hydrolyzes urea to produce ammonia (Fig. 2). Of the known homologues for the urea transporter and urease, the Barbulanympha ectosymbiont proteins are most closely related to those in “Ca. Azobacteroides pseudotrichonymphae,” and these have their closest homologues with members of the Cytophagales (Bacteroidetes) (see Fig. S3 in the supplemental material). Their genomes also contain genes involved in nitrogen fixation, including genes for nitrogenases, nitrogen regulatory proteins, and transporters specific for nitrogenase cofactors that are not typically found in other Bacteroidales (Fig. 2 and 4). These bacteria also harbor an ammonia transporter that can be found in some, but not all, Bacteroidales (Fig. 2).

FIG 4.

Genetic structure of Barb6XT contigs carrying nifH and nifH-like genes (shaded). anf nomenclature is used for genes associated with Fe-only nitrogenases. Nitrogenase genes are shown as black arrows pointing in the direction of transcription. Genes for nitrogen regulation are in medium gray, and other genes whose functions support nitrogen fixation are in dark gray. All other genes are in light gray.

The Barbulanympha ectosymbiont genomes carried two distinct nitrogenase operons, one for a molybdenum (Mo)-dependent nitrogenase and a second for an alternative iron (Fe)-only nitrogenase (Fig. 4). The Barb6XT Mo-dependent nitrogenase operon included the nifH, nifD, nifK, nifE, and nifN genes as well as two nitrogen PII signal transducer genes (glnB) (Fig. 4). An ABC transporter for molybdenum was located upstream of the nitrogenase genes, and downstream were a molybdenum cofactor biosynthesis gene and a molybdate-dependent transcriptional regulator. The Barb6XT Fe-only nitrogenase operon consisted of anfH, anfD, and anfK (nifH, nifD, and nifK homologues, respectively), two glnB ORFs, and a downstream nitrogenase accessory protein (anfO) (Fig. 4). The genes for these nitrogenase operons were also found in the Barb4, Barb6, and Barb7 assemblies, but split across multiple contigs.

The amino acid sequences of the NifH protein from these nitrogenase operons were identical, or nearly so, to either cluster II (Fe only) or cluster III-3 (Mo-dependent) NifH previously sequenced from the hindgut of Cryptocercus punctulatus (following the nomenclature of Yamada et al. [36]) (Fig. 5). They also grouped with other Bacteroidales NifH homologues, including ones from other symbionts of termite hindgut protists and recently discovered from cultivated strains but not from Prevotella bryantii B₁4 (37, 66).

FIG 5.

Maximum likelihood tree of NifH amino acid sequences. Environmental clones consisted of all sequences obtained from the hindgut of Cryptocercus punctulatus (Cp clones, accession numbers AB273412 to AB273459). Black vertical bars identify recognized clusters of NifH sequences. NifH sequences from Bacteroidales are labeled in bold, and those from Barbulanympha ectosymbionts are shaded. Only bootstrap support values greater than 70% from 1,000 replicates are shown.

The Barb6XT, Barb4, and Barb7 ectosymbiont genomes encoded an additional NifH protein that grouped within a clade of poorly understood NifH relatives (cluster IV) (Fig. 5). In the Barb6XT and Barb7 assemblies, this nifH gene was located next to homologues of nifD and nifK (Fig. 4). A second cluster IV nifH gene was found in the Barb6XT, Barb4, and Barb6 ectosymbiont genomes, closely related to the first. Like the first cluster, these NifH-related genes were nearly identical to each other and previously sequenced cluster IV NifH gene from C. punctulatus hindguts (Fig. 5).

The cluster IV NifH from the Barbulanympha ectosymbionts groups with the only NifH homologue from Bacteroides reticulotermitis, isolated from the gut of the subterranean termite Reticulitermes speratus (Fig. 5). B. reticulotermitis also has genes that encode a NifK-like protein and a truncated NifD-like protein that are located next to its nifH gene, and these proteins are closely related to the ones in the Barbulanympha ectosymbionts (see Fig. S4 and S5 in the supplemental material). Together, these proteins were distantly related to their homologues in Endomicrobium proavitum, an Elusimicrobia isolate from the hindguts of Reticulitermes santonensis recently shown to be linked to nitrogen fixation (71).

Evidence for nitrogen fixation from stable isotope labeling and NanoSIMS analysis.

To assess the likelihood that Barbulanympha ectosymbionts actively fix nitrogen, live C. punctulatus nymphs were housed in a ¹⁵N₂ atmosphere for 2 weeks, after which their hindgut contents were collected by dissection and the hindgut microbes examined using ¹⁵N/¹⁴N imaging analysis with NanoSIMS. Four Barbulanympha cells were identified by SEM for analysis, including two that were damaged during fixation such that the cell interiors were exposed adjacent to attached Bacteroidales ectosymbionts (Fig. 6). From these two exposed Barbulanympha cells, NanoSIMS analysis showed that the ectosymbionts were significantly more enriched in ¹⁵N than their host (2.46 versus 1.47 APE ¹⁵N and 1.73 versus 1.15 APE ¹⁵N, respectively; P < 0.001 for both) (Fig. 6D). The ectosymbionts on two of the intact Barbulanympha cells were also enriched (4.26 and 4.68 APE ¹⁵N) (Fig. 6D).

FIG 6.

NanoSIMS analyses of nitrogen fixation in Barbulanympha ectosymbionts and free-living bacteria. (A) Scanning electron micrograph of a damaged Barbulanympha cell that was analyzed by NanoSIMS. Scale bar is 20 μm. (B) Area of cell corresponding to the red box in panel A. Here, the interior of the host cell was exposed and host cytoplasm could be clearly differentiated from the ectosymbionts. Scale bar is 5 μm. (C) NanoSIMS image of the same area as in panel B. Colors reflect ¹⁵N enrichment relative to ¹⁴N. Black areas are where the extracted ion counts were very low (<5% of maximum) due to surface topography and were excluded from analysis. (D) NanoSIMS ¹⁵N enrichment data. Solid circles ringed in blue are the mean ¹⁵N enrichments of ion measurements from the cytoplasm of individual Barbulanympha cells, open squares are from individual ectosymbionts on the cell surface, open diamonds are from individual free-living bacteria in the hindgut, and crosses are from protists in the hindgut other than Barbulanympha. The number of individual bacterial cells measured is indicated (n). Red lines indicate the mean ¹⁵N enrichment ± the standard error for the ectosymbionts on a single host cell or for free-living bacteria. The standard error line is not visible if it is smaller than the thickness of the mean line. Enrichment data for Barbulanympha 3 and 4 were not available because the host cell was intact and the cytoplasm was not accessible for analysis. Measurement uncertainties (standard errors) are not shown. They are smaller than the data points for Barbulanympha cells, their ectosymbionts, and other protists and are 1 to 4% for the free-living bacteria.

For comparison, unidentified free-living bacteria and other protists from the hindgut were also analyzed by NanoSIMS. Approximately 90% of the bacteria had no significant ¹⁵N enrichment. Of the 33 bacteria with significant enrichment, only seven had enrichment levels reaching the range of the analyzed Barbulanympha ectosymbionts (Fig. 6D). Of the other protists examined, 75% had no significant ¹⁵N enrichment (6 of 8 cells). Two protist cells were enriched to levels similar to those in Barbulanympha but lower than the majority of the ectosymbionts (Fig. 6D).

Genetic content of the ectosymbionts from each host cell is variable and polymorphic.

The ectosymbiont genomes from each host cell were compared by clustering the ORFs into homologous gene clusters. A large proportion of the gene clusters were unique to an ectosymbiont. Most interestingly, four ORFs (TraG, TraJ, TraO, and TraN) that are part of a Bacteroides conjugative transposon were found only from the Barb7 contigs, but the large majority of unique clusters were hypothetical and did not contribute to the metabolic and functional analysis of the ectosymbionts (see Fig. S6 in the supplemental material). Nevertheless, for the unique clusters, a gene belonging to a Bacteroidetes was the best hit for 71 to 85% of the hits to GenBank's nonredundant (nr) nucleotide database (see Fig. S6), indicating that potential contamination of contigs from other organisms was low. Approximately one-third of the unique clusters (22 to 39%) were not truly unique, but homologues in the other genomes were not detected because the ORFs were truncated or misannotated (see Fig. S6).

The ORFs found in common among all four sets of ectosymbiont contigs shared on average 94.9% nucleic acid similarity. Most of the single nucleotide polymorphisms (SNPs) were unique to the ectosymbionts from a single Barbulanympha cell, indicating that distinct ectosymbiont strains were harbored by each host cell (see Fig. S7 in the supplemental material). The SNPs were largely synonymous and did not result in amino acid changes (data not shown). Gene order also varied on the ectosymbiont contigs, which was confirmed by PCR, suggesting that rearrangement possibly mediated by transposases is also playing a significant role in the evolution of these bacterial genomes (see Fig. S8 and S9 in the supplemental material). In contrast, there were far fewer polymorphisms among the ectosymbiont sequences from a single Barbulanympha cell. The sequencing reads from a single Barbulanympha cell originated from hundreds of ectosymbionts on its cell surface, but there was little variation in sequence, especially of the ORFs common to all four ectosymbionts (see Fig. S7). The lack of variation in the sequencing reads suggests that the ectosymbionts from a single host cell were mostly clonal.

DISCUSSION

The Barbulanympha ectosymbiont genome is not reduced.

The Barbulanympha ectosymbiont genomes were not reduced in size or gene content, nor were they AT biased, despite the fact that the Barbulanympha-Bacteroidales symbiosis is considered highly specific, vertically transmitted, and obligate (13, 24, 32). Under these conditions, such genomic features are typical for endosymbionts (19–21), including those of Bacteroidales (16), because the symbionts undergo population bottlenecks during vertical transmission resulting in relaxed selection and genetic drift. Not all endosymbionts have reduced genomes (41), but this raises the question of whether the genomes of ectosymbionts evolve in a fundamentally different way.

Much like endosymbionts, the extracellular bacteria associated with the guts of plataspid and acanthosomatid stinkbugs (Plataspidae and Acanthosomatidae) exhibit AT-biased, reduced genomes and accelerated rates of molecular evolution (72, 73). These extracellular symbionts are harbored in specialized gut crypts that are isolated from the main gut tract, similar to the stable intracellular environments of endosymbionts, and their deposition on the surface of eggs results in vertical transmission when the symbionts are probed or ingested by newborn nymphs (10, 73). Thus, extracellular symbionts can evolve similarly to endosymbionts.

In contrast, like the Barbulanympha ectosymbionts, extracellular symbionts of lumbricid earthworms (colonizing the lumens of nephridia) also do not have reduced, AT-biased genomes even though the association is vertically transmitted, mutualistic, highly specific, and evolutionarily ancient (∼100 million years) (74, 75). These symbionts are deposited in egg capsules and transmitted to the nephridia of developing worms through a recruitment canal (76). Due to this vertical transmission, they have likely undergone population bottlenecks similarly to endosymbionts, as shown by their accelerated evolutionary rates, but genome reduction has not evolved (75, 77). These symbionts also have dynamic genomes where an expansion of mobile elements appears to have mediated genome rearrangements. With the large number of transposases, variable gene content, and preliminary PCR evidence of rearrangements, we also predict that the Barbulanympha ectosymbiont genomes are dynamic.

The lack of genome reduction and active genome rearrangements in the extracellular symbionts of lumbricid earthworms are attributed to their interaction with mixed microbial communities and multiple environments rather than the genetically isolated and stable environments of intracellular symbionts. Diverse and fluctuating environments may select against the loss of genes that are required for survival under these conditions and provide a milieu for genetic exchange and recombination with other microbes, which would counteract the deleterious effects of bottleneck-induced genetic drift (77).

The interaction of the Barbulanympha ectosymbionts with the mixed microbial community in the hindgut environment may also explain their lack of genome reduction, even though they are obligate, specific, vertically transmitted symbionts. Both genomic and morphological data indicate that they have retained their cell wall, outer membrane, and outer membrane proteins, effectively maintaining a boundary for mediating interactions with the hindgut environment (13) (Fig. 2 and 3). Efflux transporters were also retained providing a defensive capacity for living in this mixed community (Fig. 2 and 3). Hence, the extracellular symbiotic lifestyle itself with population bottlenecks does not necessarily result in converging evolutionary trajectories with endosymbionts. Rather, the extent of genetic isolation and the stability of a symbiont's environment likely play major roles in the evolution and genetic content of their genomes.

Specificity of the Barbulanympha-Bacteroidales symbiosis.

The selective pressures faced by Barbulanympha ectosymbionts and the evolutionary outcomes observed in their genome structure are most likely the result of their niche as extracellular symbionts attached to their protist host and not due to a free-living stage within the insect gut. Barbulanympha is usually covered in rod-shaped ectosymbionts, and morphologically identical bacteria (possibly the same bacteria as the ectosymbionts) have also been observed intracellularly in vesicles but are much less abundant (13, 24). The ectosymbionts appear to be attached to Barbulanympha cells through specialized, electron-dense structures in the host cytoplasm and a thickened glycocalyx (13, 24). The proteins responsible for this interaction are not known but could include the numerous outer membrane proteins, receptors, and a possible adhesin protein in the ectosymbiont. The density of ectosymbionts on Barbulanympha cells varies, possibly decreasing during Barbulanympha cell division and encystment (13), but Barbulanympha cells without ectosymbionts have not been observed. Repopulation of the Barbulanympha cell surface following division or emergence from cysts has been hypothesized to come from symbionts that have remained on the cell through its life stages or from symbionts that were retained in vesicles intracellularly (13). These ultrastructural data suggest that the symbiosis is obligate, and the symbionts are vertically transmitted during Barbulanympha cell division and not acquired from the hindgut environment, although the possibility of a free-living stage cannot be completely rejected. Barbulanympha itself is vertically transmitted from adult insects to nymphs together with other hindgut symbionts through proctodeal trophallaxis (the consumption of hindgut fluids directly from the rectal pouch of the donor) (78, 79).

Protist-Bacteroidales symbioses have evolved many times in the hindguts of lower termites and Cryptocercus cockroaches, and different protist lineages independently acquired symbionts from the pool of bacteria in the hindgut (32). But once acquired, the protist host and symbiont appear to have cospeciated in many cases, indicating the high degree of specificity between hindgut protists and their Bacteroidales symbionts (29, 31). As demonstrated by the near identity of 16S rRNA sequences from Barbulanympha isolated from different collections of C. punctulatus cockroaches (Fig. 1), Barbulanympha consistently associates with a specific lineage of Bacteroidales. Molecular data were not collected from the cells analyzed by NanoSIMS, but there is little doubt these symbioses are consistent with those investigated for genomics.

Bacteroidales as nitrogen fixers and recyclers.

Nitrogen fixation has been hypothesized from molecular and genomic data to be a key process in termite gut symbioses (16, 35–37, 41). Our NanoSIMS analysis of in vivo ¹⁵N-labeled hindgut microbes provides direct imaging evidence of a specific hindgut microbe playing a role in nitrogen fixation in its natural environment and represents a significant advance for these uncultivated microbes. The ectosymbionts were significantly more enriched in ¹⁵N than their host Barbulanympha and were much more enriched than the majority of free-living bacteria in the hindgut (Fig. 6). Barbulanympha is likely enriched in ¹⁵N because it receives fixed nitrogen products from its ectosymbionts (likely in the form of ammonia, but this is not known). Other protists in the hindgut are not similarly enriched, but those that are might also harbor nitrogen-fixing symbionts. Approximately 2% of the free-living bacteria were as enriched as the Barbulanympha ectosymbionts, providing evidence that free-living bacteria also fix nitrogen in the C. punctulatus hindgut. However, most of the bacteria (∼80%) were not significantly enriched (Fig. 6).

Bacteroidales are often associated within the guts of animals and are typically known as degraders of polymeric carbohydrates (80), but recent genome sequencing has revealed the potential for nitrogen fixation in several members (66). It is also becoming increasingly clear that most Bacteroidales endo- and ectosymbionts of protists in the hindguts of termites and cockroaches are likely diazotrophs (16, 37). The hindgut of Cryptocercus punctulatus harbors microorganisms with a diversity of nifH genes (36), and we show here that four of these are from the ectosymbionts of Barbulanympha encoding an Fe-only nitrogenase subunit (cluster II), a Mo-dependent nitrogenase subunit (cluster III-3), and 2 cluster IV NifH proteins (Fig. 4).

The Barbulanympha ectosymbiont is the only known Bacteroidales genome with homologues from three different NifH clusters. The presence of both Fe-only and Mo-dependent nitrogenases is thought to maintain nitrogen fixation depending on the availability of Mo cofactors. However, in termite guts and in Bacteroidales symbionts encoding homologues to both Fe-only and Mo-dependent NifH, Fe-only NifH (AnfH) was preferentially expressed despite providing additional molybdenum (35, 37). Cluster IV NifH was not thought to contribute to nitrogen fixation, but Endomicrobium proavitum (belonging to the Endomicrobia and a free-living termite hindgut microbe) fixes nitrogen while only encoding a cluster IV NifH, and also NifD-like and NifK-like proteins (71). Barbulanympha ectosymbionts have multiple, diverse nitrogenase operons, suggesting that the ectosymbionts have several means to achieve nitrogen fixation. The NanoSIMS analysis demonstrates that the ectosymbionts do fix nitrogen, but the specific nitrogenase genes that are responsible for this function and the expression and regulation of these genes have not been determined.

Bacteroidales with nitrogen fixation genes belong to different lineages based on 16S rRNA gene phylogeny, but their NifH proteins cluster together (Fig. 4 and 5), suggesting that their nitrogenase genes are ancestral and have generally evolved vertically but were lost in many Bacteroidales (66). The diversity of nifH homologues and associated nitrogenase genes in the Barbulanympha ectosymbiont suggests that a common ancestor possessed a diversity of nitrogenases that were subsequently lost or transferred among the Bacteroidales. Alternatively, the nitrogenases may have been independently acquired in different Bacteroidales ancestors that were transferred among the Bacteroidales, with 3 types accumulating in the Barbulanympha ectosymbiont lineage. The two cluster IV nifH genes in the Barbulanympha ectosymbionts likely arose through duplication. Notably, the NifH from Prevotella bryantii B₁4 did not cluster with the majority of the Bacteroidales NifH proteins, suggesting that P. bryantii B₁4 nifH has a different origin and that the occurrence of nifH genes in the Bacteroidales is not entirely due to common ancestry. Deviations from a strict vertical evolution of nitrogenase genes in the Bacteroidales have also been noted previously (66).

By providing bioavailable nitrogen, nitrogen fixation in the Bacteroidales, and perhaps also urea transport and metabolism, has likely been key to their evolution as protist symbionts in the hindguts of wood-eating termites and cockroaches. Low nitrogen availability is the most plausible driving force behind these protist-Bacteroidales symbioses because wood, which is consumed by many of the large protists in the hindguts, is nitrogen poor. The prevalence of nitrogen fixation by bacterial symbionts of protists supports this hypothesis, but the amount of nitrogen supplied to protists by their symbionts, the mechanism, the type of molecule supplied, and the dependency of the protists on this nitrogen supply are not known. Free-living hindgut bacteria probably also fix nitrogen (38–40), but most of the nitrogen fixation activity in the hindguts is likely performed by bacterial symbionts of protists (37, 41), a conclusion supported by our analysis of free-living bacteria in the hindguts of Cryptocercus (Fig. 6D). Consequently, these protist-bacterium symbioses are key to the hindgut ecosystem, which functions to digest wood and supply nutrition to their insect hosts.

Conclusions.

The combination of culture-independent genomics and NanoSIMS analyses provides a powerful set of complementary tools for investigating the biology of uncultivable microbes. We used these tools to study the Bacteroidales ectosymbionts of Barbulanympha, providing new insights into the functional interactions and evolution of ectosymbiosis. Like Bacteroidales that are protist endosymbionts, Barbulanympha ectosymbionts are diazotrophs and probably also recycle nitrogenous compounds. The ecological pressures faced by these ectosymbionts, however, clearly differ from those for endosymbionts, which reside intracellularly, resulting in divergent evolutionary outcomes. The dynamic environment experienced by ectosymbionts is in sharp contrast to the stable intracellular environment of endosymbionts, which we suggest favors gene retention and genome recombination. However, the ecological and evolutionary conditions that result in intracellular symbiosis as opposed to extracellular symbiosis remain unclear. It is tempting to speculate that the Barbulanympha ectosymbionts might eventually evolve into endosymbionts, as may have occurred with the spirochete endosymbionts of the termite gut-dwelling protist Eucomonympha (41), but the ectosymbionts are clearly adapted to interact with a diverse extracellular environment, and their genomes, though not reduced, are in a process of dynamic evolution.

Funding Statement

This work was supported by a grant (227301) to Patrick Keeling from the Natural Sciences and Engineering Research Council of Canada (NSERC). NanoSIMS work was supported by an LLNL Laboratory Directed Research and Development grant (011-LW-039) to Kevin Carpenter. Peter Weber at LLNL was funded in part by DOE Genome Sciences Program grant SCW1039. Vera Tai was supported as a Global Scholar with the Canadian Institute for Advanced Research (CIFAR) and through a Postdoctoral Fellowship from NSERC. CIFAR supports Patrick Keeling as a Senior Fellow and Steve Perlman as a Fellow of the Integrated Microbial Biodiversity Program.