Genetic differentiation among isolates of Teredinibacter turnerae, a widely occurring intracellular endosymbiont of shipworms

Abstract

Teredinibacter turnerae is a cultivable intracellular endosymbiont of xylotrophic (wood-feeding) bivalves of the Family Teredinidae (shipworms). Although T. turnerae has been isolated from many shipworm taxa collected in many locations, no systematic effort has been made to explore genetic diversity within this symbiont species across the taxonomic and geographic range of its hosts. The mode of symbiont transmission is unknown. Here we examine sequence diversity in fragments of six genes (16S rRNA, gyrB, sseA, recA, rpoB, and celAB) among 25 isolates of T. turnerae cultured from 13 shipworm species collected in 15 locations in the Atlantic, Pacific and Indian Oceans. While 16S rRNA sequences are nearly invariant among all examined isolates (maximum pairwise difference <0.26%) variation among examined protein coding loci is greater (mean pairwise difference 2.2-5.9%). Phylogenetic analyses based on each protein-coding locus differentiate the 25 isolates into two distinct and well-supported clades. With five exceptions, clade assignments for each isolate were supported by analysis of alleles of each of the five protein coding loci. These exceptions include (1) putative recombinant alleles of the celAB and gyrB loci in two isolates (PMS-535T.S.1b.3 and T8510), suggesting homologous recombination between members of the two clades, and (2) evidence for a putative lateral gene transfer event affecting a second locus (recA) in three isolates (T8412, T8503 and T8513). These results demonstrate that T. turnerae isolates do not represent a homogeneous global population. Instead they indicate the emergence of two lineages that, although distinct, likely experience some level of genetic exchange with each other and with other bacterial species.

Introduction

Teredinibacter turnerae is a gammaproteobacterium of the order Alteromonadales known to occur as an endosymbiont of wood-boring bivalves of the family Teredinidae (commonly known as shipworms). Shipworms are among the most important wood consumers in marine environments [1]. Although capable of filter feeding [2], these bivalves are the only marine animals known to sustain normal growth and reproduction with wood as their sole source of particulate food [3]. Shipworms are ecologically and economically important, acting both as the principal mineralizers of wood in marine environments and as costly nuisance species, causing extensive damage to wooden piers, vessels, fishing equipment, and other man-made structures [4]. Despite their economic importance, and the fact that they have been part of the scientific literature since their description by Sellius in 1753 [5], this family remains poorly studied and poorly understood. The taxonomy of the family Teredinidae was most recently reviewed by Turner in 1966 [1]. This work, which recognizes only 14 of 42 historically proposed genera and 66 species from among nearly 300 historical species names, forms the basis of the classification scheme applied here [6] but also underscores the chaotic state of the systematics of this family. Indeed, recent molecular phylogenetic analyses show major taxonomic divisions within the family to be nonmonophyletic [7]. The geographic distribution of Teredinidae is also poorly understood. Many species are distributed over vast oceanic areas “apparently limited only by temperature, salinity and [means of] transportation [8]. Current-driven dispersal likely contributes to the frequent transport of long-lived planktonic larvae over entire ocean basins [9]. Adults also may be frequently transported over great distances by drifting wood and by wooden vessels. Thus, range boundaries of many Teredinidae may be highly dynamic in time and space and are only poorly characterized.

The mode of wood digestion by shipworms is also largely unknown, but intracellular symbionts have been implicated in this process [10]. Most described wood-eating animals rely on cellulolytic enzymes produced by extracellular microbes in the gut and/or endogenous nuclear-encoded cellulases for digestion of wood [4, 11]. Shipworms lack such well-developed microbial communities in their gut [12], but harbor dense endosymbiotic microbial communities within a specialized tissue of the gills referred to as the “Gland of Deshayes” [13]. The presence of bacterial endosymbionts in shipworm gills was first demonstrated by Popham and Dixon in 1973 using transmission electron microscopy (TEM) [14, 15]. Various putative functions were attributed to these bacteria, including production and/or uptake of essential amino acids [16, 17]. However, the subsequent discovery and in vitro cultivation of a bacterium, called Teredinibacter turnerae [18], from gill homogenates from numerous shipworm species, suggested alternative functions for these intracellular symbionts [10]. When grown in pure culture, T. turnerae secretes cellulase (endo-1, 4-β-D glucanase) [19-21] and other key enzymes that degrade wood lignocellulose. It also fixes nitrogen when grown micro-aerobically [10]. Based on this information, it was proposed that T. turnerae is the intracellular endosymbiont previously observed by Popham and Dixon and that it produces enzymes, including cellulases and nitrogenase, that may contribute to the host's ability to survive on an indigestible and nitrogen-deficient diet of wood [10, 20, 22].

The presence of T. turnerae in gill bacteriocytes of the shipworm L. pedicellatus was subsequently confirmed using culture independent 16S rRNA methods including fluorescence in situ hybridization (FISH) with 16S rRNA directed oligonucleotide probes [23] but these studies also revealed that that T. turnerae is not alone in shipworm gills. Instead it may constitute one member of a community that may also include several additional as yet uncultivated but closely related symbiont phylotypes [24, 25]. For example, in the shipworm Lyrodus pedicellatus, five symbiont phylotypes were detected, one of which corresponded to T. turnerae (on average accounting for 11.5 ± 8.2% of total 16S rRNA copies), and all of which were shown to form a closely related (99.8-91.2% identical in 16S rRNA sequence) monophyletic group that excluded other known free-living bacteria based on analyses of 16S rRNA genes [12, 24, 25]. These symbionts have been shown to fix nitrogen that is subsequently utilized by the host [26], and have been proposed to produce cellulolytic enzymes that contribute to the host's ability to digest wood [10]. In addition, T. turnerae has been shown to produce antimicrobial compounds that may help it to defend its intracellular niche and to structure the microbial communities associated with the host [27]. The mode of transmission of T. turnerae remains unknown although evidence for vertical transmission has been reported for another closely related but as yet uncultivated shipworm gill endosymbiont species [28].

T. turnerae differs from other shipworm endosymbionts, and other bivalve gill endosymbionts, in that it has been described based on axenic culture and that it has been isolated from many shipworm specimens, species and genera collected in many locations around the world [10, 18]. Previous phylogenetic analyses of six T. turnerae isolates from diverse host species, based on reverse transcribed 16S rRNA sequences, detected no sequence variation among isolates [23]. These results raised the question of whether T. turnerae represents a single, largely undifferentiated, symbiont population with worldwide distribution, or whether as yet undetected genetic structure exists among isolates. To explore this question, we took advantage of, and added to, a unique existing collection of T. turnerae isolates obtained from a broad cross section of the taxonomic and geographic range of its host species between 1979 and 1986 by John B. Waterbury (WHOI) and Ruth D. Turner (Harvard University). We examined sequences of multiple gene loci including 16S rRNA and five protein-coding loci in twenty-five T. turnerae isolates. We reasoned that the extent and patterns of sequence variability at these loci might provide clues with regard to specific relationships among symbionts, incidence of genetic exchange among symbionts, and patterns of host-symbiont sorting. While the opportunistic sampling performed here is not suitable for rigorous analysis of host and symbiont biogeography or cophylogeny, nor to resolve the as yet unknown mode(s) of symbiont transmission, it is sufficient to place constraints on, and therefore to provide context for, our understanding of this little studied and poorly understood symbiosis.

Materials and Methods

Isolation of T. turnerae

Isolates T7901, T7902, T7903, T8401, T8402, T8412, T8415, T8503, T8508, T8509, T8510, T8513, T8601 and T8602, were obtained from a collection of isolates cultured by J. Waterbury (WHOI) between 1979 and 1986 and subsequently deposited to the Ocean Genome Resource (www.oglf.org). These isolates were cultivated as in [10, 18]. Briefly, gills were removed from freshly dissected shipworm specimens, rinsed in sterile 0.22μm filtered seawater (FSW) and gently homogenized by hand in a glass Dounce homogenizer in ≥10 volumes of fresh FSW. One ml of homogenate was then used to inoculate soft agar cultures in shipworm basal medium (SBM) containing 0.2% agar and 0.2% powdered cellulose (Sigmacell cellulose Type 101) as the sole potential sources of organic carbon (SBM+S). Ten serial dilution steps were performed on each homogenate with ten-fold dilution (v/v) at each step. Individual colonies were obtained from the highest dilution in which growth was observed by streaking onto SBM agar plates with 0.2% Sigmacell and 5 mM ammonium chloride added as sole sources of organic carbon and combined nitrogen respectively (SBM+S+N plates). Plates were incubated at 22-30°C until colonies were observed (4-7 days). Individual colonies were picked and re-streaked on fresh plates at least twice, then used to inoculate liquid cultures. Liquid cultures for DNA extraction were grown from single colonies in 25ml of SBM+S+N in 250 ml Erlenmeyer flasks in a shaker incubator at 22-30°C for 5-10 days, shaken at 100 rpm. Stocks for long-term storage were prepared by adding 0.5 ml of 40% glycerol to 0.5 ml of 5 day liquid culture and freezing at −80°C. All other isolates were prepared by the same method except that the soft agar culture and serial dilution steps were omitted. Instead, homogenates were directly streaked on SBM+S+N plates.

Host and isolate selection

The twenty-five T. turnerae isolates examined in this investigation were isolated from thirteen host species, selected to represent a broad cross-section of the phylogenetic diversity and geographic range of the bivalve family Teredinidae. These included representatives of two of three extant subfamilies (Teredininae and Bankiinae) and seven of fourteen genera [1, 29]. Collection sites included locations in the Atlantic, Pacific and Indian oceans and the coasts of four continents (North and South America, Asia and Australia). Where possible, isolates were obtained from multiple representatives of single host species collected both in closely situated and distantly separated locations. Isolates T0609 and T0611 and isolates PMS-509L.S.1a.4 and PMS-509L.S.1a.6 were obtained from single specimens of the species L. pedicellatus and Teredo clappi, respectively. Isolates T0611 and T0609 were obtained from a specimen of L. pedicellatus from the same continuously maintained captive breeding colony as was T7902 twenty-seven years earlier. See Table 1 for a complete list of isolates, hosts and locations.

Table 1.

Isolates, hosts, locations, loci and clades

Symbiont	Host					Clade
Isolate1	Genus	Species	Location	Latitude2	Longitude2	gyrB	sseA	celAB	recA	rpoB
T8602	Dicyathifer	manni	Townsville, QLD, Aus.	S 19.276313	E 147.057843	I	I	I	I	I
PMS-539Y.S.1a.1	Dicyathifer	manni	Bohol, Isl. Philippines	N 10.102222	E 124.334722	I	I	I	I	I
T8402	Teredora	malleolus	Floating wood	N 38.30667	W 69.59333	I	I	I	I	I
T8508	Neoteredo	reynei	Ubatuba, Brazil	S 23.491273	W 45.163853	I	I	I	I	I
T8509	Nausitora	fusticula	Ubatuba, Brazil	S 23.491273	W 45.163853	I	I	I	I	I
T7901	Bankia	gouldi	Beaufort, NC, USA	N 34.717373	W 76.671983	I	I	I	I	I
T8415	Bankia	gouldi	Fort Pierce, FL, USA	N 27.480633	W 80.309673	I	I	I	I	I
T0609*	Lyrodus	pedicellatus	Long Beach, CA, USA	N 33.761383	W 118.172813	II	II	II	II	II
T0611*	Lyrodus	pedicellatus	Long Beach, CA, USA	N 33.761383	W 118.172813	II	II	II	II	II
T7902	Lyrodus	pedicellatus	Long Beach, CA, USA	N 33.761383	W 118.172813	II	II	II	II	II
PMS-554M.S.1a.4	Lyrodus	pedicellatus	Cataban Isl., Philippines	N 10.233333	E 124.383333	II	II	II	II	II
PMS-574K.S.1a.1	Lyrodus	pedicellatus	Cataban Isl., Philippines	N 10.233333	E 124.383333	II	II	II	II	II
T8503	Lyrodus	floridanus	Key Biscayne, FL, USA	N 25.689763	W 80.163803	II	II	II	II/LGT3	II
T8412	Lyrodus	bipartitus	Jim Isl., Fort Pierce, FL	N 27.476944	W 80.311944	II	II	II	II/LGT3	II
T8513	Teredo	navalis	Sao Paulo, Brazil	S 23.819923	W 45.405173	II	II	II	II/LGT3	II
T7903	Teredo	navalis	Woods Hole, MA, USA	N 41.523403	W 70.672283	II	II	II	II	II
PMS.683H.S.1a.9	Teredo	clappi	Cataban Isl., Philippines	N 10.233333	E 124.383333	II	II	II	II	II
PMS.535T.S.1b.3	Teredo	clappi	Cataban Isl., Philippines	N 10.233333	E 124.383333	II	II	I / II	II	II
PMS.509L.S.1a.4**	Teredo	clappi	Bohol Isl., Philippines	N 10.307778	E 124.399167	II	II	II	II	II
PMS.509L.S.1a.6**	Teredo	clappi	Bohol Isl., Philippines	N 10.307778	E 124.399167	II	II	II	II	II
PMS.517Y.S.1a.10	Teredo	mindanensis	Cataban Isl., Philippines	N 10.233333	E 124.383333	II	II	II	II	II
PMS.691W.S.1a.3	Teredo	mindanensis	Cataban Isl., Philippines	N 10.233333	E 124.383333	II	II	II	II	II
T8401	Bankia	rochi	Karachi, Pakistan	N 24.841073	E 66.910693	II	II	II	II	II
T8510	Bankia	rochi	Sao Sebastiao, Brazil	S 23.819923	W 45.405173	I / II	II	II	II	II
T8601	Bankia	australis	Brisbane, QLD, Aus.	S 27.786243	E 153.461843	II	II	II	II	II

¹All isolates were obtained from separate host specimens except where otherwise indicated.

²Latitude and longitude are estimated with ~10m accuracy except where otherwise indicated

³LGT = putative lateral gene transfer

^*Estimated at ~10km accuracy

^**T0609 and T0611 were isolated from the same specimen of L. pedicellatus

^***PMS.509L.S.1a.4 and PMS.509L.S.1a.6 were isolated from the same specimen of T. clappi.

Selection of gene loci

Regions of six genes where selected for sequence analysis. These included a 1,125 bp region of the16S rRNA gene (aligning with E. coli positions 227-1351), a highly conserved structural RNA widely used as a phylogenetic marker for bacteria, regions of four protein coding genes; rpoB (4,083 bp), gyrB (2,427 bp), recA (1,029 bp), and sseA (846 bp), that are typically present in a single copy and that may be considered as components of core metabolism, and celAB (1,011 bp) a single copy bifunctional glycosyl hydrolase that may be considered as a component of specialized or peripheral metabolism. Based on the available genome sequence for one isolate (T7901) the selected loci are widely distributed around the single circular chromosome, reflecting no regional bias in distribution (Figure S1).

Nucleic acid purification, PCR amplification, and sequencing

Bacterial genomic DNA was extracted from 10 ml aliquots taken from 5-day old 25 ml liquid cultures (SBM+S+N). Cells (and some undigested residual cellulose powder) were pelleted by centrifugation (10,000 × g, 30 minutes, 4°C). Genomic DNA was extracted from the pellet using the DNeasy Blood & Tissue Kit (Qiagen) according to the manufacturer's suggested protocol. Gene fragments were amplified by polymerase chain reaction (PCR) using the mixture: 20-50 ng template DNA, 12.5 μl 2X Phusion High-Fidelity PCR Master Mix (Finnzymes), 0.5 μM forward primer, 0.5μM reverse primer, 7.5 μl sterile deionized water, for a total reaction volume of 25 μl. Thermal cycling parameters were as follows: denaturation; (95°C, 4 min, one cycle); amplification; 95°C, 30 sec, primer-dependent annealing temperature (see Table 2), 30 sec, 72°C, 60 sec, 30 cycles, extension; 72°C, 300 sec, one cycle. The PCR primer sequences are shown in in Table 2. The size and homogeneity of amplified products were evaluated by electrophoresis (1.2% agarose). Amplified products were purified using QIAquick PCR Purification Kit (Qiagen). Amplicons were either directly sequenced or cloned into pCR-Blunt II-TOPO (Invitrogen, Carlsbad, CA) vector and transformed into Escherichia coli cells (C3019, NEB). Transformed cells were screened for inserts and plasmids were purified using the Qiagen Miniprep kit according to manufacturer's instructions. Clones were selected and sequenced using both vector and gene specific primers (Table 2). For direct sequencing, PCR amplicons were purified using the QIAquick kit according to the manufacturer's suggested protocol and sequenced using gene specific primers (Table 2). All samples were bi-directionally sequenced on an ABI 3130xl Genetic Analyzer using ABI Prism BIG DYE Terminator v3.1.

Table 2.

Amplification and sequencing primers

Locus	Product	Forward primer sequence (5′ to 3′)a	Reverse primer sequence (5′ to 3′)a	Annealing tempa	Expected amplicon size (bp)
rrsA, B, and C	Small subunit (16S) ribosomal RNA	27F - AGAGTTTGATCMTGGCTCAG 530Fb - GTGCCAGCMGCCGCGG 1055Fb -ATGGCTGTCGTCAGCT	1492R - TACGGYTACCTTGTTACGACTT 800Rb - TACCAGGGTATCTAATCC 1101Rb - AGGGTTGCGCTCGTT	56°C	1495
rpoB	DNA-directed RNA polymerase β-subunit	rpoB-F1 - ATGACAGACAACGGTACWTTCGTWRTBAAYG rpoB-F2b - GAAGANGGCCGTCGTATTACTGC	rpoB-R1 - GTGAGTCGGGTGTACGTCWCGYACYTC rpoB-R2b - GGCGCATCATGCGGTARATYTCDACCAR	61°C	1277
gyrB	DNA gyrase β-subunit	gyrB-F - CATGCTGGCGGTAAATTCGA	gyrB-R - TCTACGTCCGCATCCGTCAT	60°C	1211
recA	DNA recombinase A	recA-F1 - GCCAGTTTGGTAAAGGCACBGTBATGCGYM recA-F2b - GGTGCGTTCAGGCGCARTHGAYGTKYT recA1-Fbc - CGTATGGGTGACAAAGAGCTGGAAG	recA-R1 - CGCCGTTGTAGCTATACCARGCRCCV recA-R2b - CGCAGCGCTTGCGACATYARTCGMGCY recA2-Rb,c- GAGCTGCAACAGAGTCAATAATTACA	63°C 60°C	831 1150
sseA	Thiosulfate/3-mercaptopyruvate sulfurtransferase	sseA-F - ATGGCTAATTCACCCCTGGTG	sseA-R - TTAGGTACTGGTGGCGATGGG	60°C	849
celAB	Bifunctional glycoside hydrolase	celAB-F - CCACCTCTGCAGCTTTCGCGG	celAB-R - GCCGTTACCGATTGTGGTGCC	61°C	941
	Cloned PCR amplicons	M13-Forward(-20) - GTAAAACGACGGCCAG	M13-Reverse - CAGGAAACAGCTATGAC

^aPCR conditions: initial denaturation for 3 min at 98°C; 35 cycles of 30 sec at 98°C, 30 sec at annealing temperature, 1 min at 72°C; final extension for 7min at 72°C; hold at 4°C.

^bInternal sequencing primer

^cInternal primer pair specific to the recA₁ and recA₂alleles of isolates T8412, T8503, and T8513. Complementary sites of recA primers are shown in Figure S2

Phylogenetic Analysis

Nucleotide sequences for each gene fragment were assembled and edited using Geneious (v. 6.0.5) (Biomatters, NZ, available from http://www.geneious.com). Alleles of each locus were aligned using MAFFT [30] implemented in Geneious (v. 6.0.5) under default parameters (alignment strategy selected automatically; scoring matrix 200PAM/k=2; gap open penalty 1.53; and offset value 0.123; and trimmed to equal length (16S rRNA, 1125 bp; gyrB, 1,053 bp; recA, 489 bp; rpoB, 1,020 bp; celAB, 804 bp; sseA, 591 bp). GenBank accession numbers are listed in Table S1. Sequence divergence and base composition of the aligned datasets were calculated using DnaSP version 5.10 [31]. Best-fit models of evolution and parameter value estimates for single gene and concatenated datasets were identified by Akaike information criterion (AIC) through jModelTest 2 [32] (GTR+I+Γ for 16S rRNA, gyrB and concatenated dataset; SYM+ Γ for recA and rpoB; SYM+I for celAB; HKY85+I for sseA.) The partition homogeneity test (p = 0.01) [33] was used to determine compatibility of the data prior to concatenation. Phylogenetic analyses were performed by employing both Bayesian inference (MrBayes) [34, 35] and maximum likelihood (PhyML) [36] methods implemented in Geneious version 6.0.5. Bayesian-inference analyses were executed using 5 million generations, subsampling every 2000 generations with a burn-in of 20%. The sump command in MrBayes was used to confirm convergence on similar optimal log likelihood scores and tree topologies in each analysis. Maximum likelihood trees were constructed using a combination of Nearest Neighbor Interchanges (NNIs) and Subtree Pruning and Regrafting (SPR) algorithms. A total of 100 bootstrap replicates were used to construct maximum likelihood trees. Trees based on amino acid sequences were inferred using the maximum likelihood algorithm implemented in Molecular Evolutionary Genetics Analysis (MEGA) software version 5.05 [37] employing Jones-Taylor-Thorton (JTT) amino acid substitution model with 5 discrete gamma categories for a total of 1000 bootstrap replicates. Congruence of host and symbiont phylogenies was analyzed using the randomization test implemented in TreeMap v.2 [38] comparing tree topologies inferred for symbionts as described above and hosts as described in [7]. To examine the relationships of the T. turnerae isolates to other gammaproteobacteria, additional phylogenetic analyses were performed as described above using 16S rRNA, gyrB, recA, and rpoB with inclusion of homologous reference sequences from related gamma proteobacterial species. BLAST searches of Genbank non-redundant nucleotide database failed to identifiy compelling homologues of celAB and sseA in bacterial species other than T. turnerae and so these were omitted from this analysis. E. coli sequences were specified as outgroups for gyrB, recA, rpoB and 16S rRNA in Bayesian inference analyses.

Detection of recombination

Individual gene datasets were screened for evidence of recombination using GENECONV [39], BootScan [40], MaxChi [41], Chimaera [42], and 3Seq [43] methods implemented in Recombination Detection Program (RDP) version 4.16 [44] with a P-value of 0.01, using a Bonferroni correction [45] to compare test results. For MaxChi, Chimaera, and RDP, a window size of 60 nucleotide positions was used to scan the sequences for recombination; while for Bootscan a step size of 20 was selected and the Felsenstein 1984 model was employed with 1000 bootstrap replicates. G-ENECONV analysis was performed with G-scale (mismatch penalty) set to 1.0.

Analysis of selection

Evidence for positive or negative selection was assessed by analysis of nonsynonymous and synonymous substitutions within the 5 protein-coding loci using SLAC, FEL, REL and FUBAR [46] methods implemented on the Datamonkey server (http://www.datamonkey.org) [47, 48]. Recombinant alleles identified by methods within RDP were removed from the dataset. The best-fit nucleotide substitution model was reestimated using HyPhy (gyrB, GTR 011020; rpoB, TN93; celAB, 010011; recA and sseA, HKY85)[49]. Positively and negatively selected sites were determined using a cutoff P-value of 0.05 for SLAC and FEL tests; Bayes Factor of 100 for REL; and a posterior probability of 0.95 for FUBAR. Results from all the four tests were then compared using the integrative selection tool within Datamonkey to determine consensus among the different methods.

Results

Sequence variation among examined loci

Partial gene sequences (489 to 1,125 nucleotides) were determined for five protein-coding gene loci (rpoB, recA, gyrB, sseA and celAB) from the genomes of 25 T. turnerae isolates and one structural RNA (16S rRNA) from 21 of the aforementioned isolates. All isolates examined were nearly identical with respect to 16S rRNA sequence. Of 1,125 nucleotide positions in the aligned 16S rRNA sequence set, only three were observed to be variable. The phylogenetic position of T. turnerae isolates with respect to other gammaproteobacteria is shown in Figure S2. The greatest pairwise difference observed among isolates in examined 16S rRNA sequences was 3/1,125 or 0.26% (data not shown). Because 16S rRNA sequences were nearly invariant, this locus was excluded from further analyses. The protein-coding loci showed considerably greater variation with mean pairwise differences ranging from 2.2-5.9% (Table 3). With the exception of three recA alleles in three strains (T8412, T8503 and T8513; see Recombination and Lateral Gene Transfer), no length variation was observed among PCR amplicons derived from the examined alleles.

Table 3.

Mean interclade and intraclade distances (substitutions/nucleotide position calculated from ungapped nucleotide alignments¹).

Gene	All pairwise distances1,4,	Intraclade2		Interclade3
		Clade I	Clade II	Clade I to Clade II
gyrB	0.059	0.020	0.014	0.118
rpoB	0.047	0.003	0.006	0.103
recA	0.022	0.014	0.007	0.041
celAB	0.027	0.003	0.007	0.054
sseA	0.038	0.011	0.010	0.078
All loci (concatenated)	0.043	0.010	0.009	0.086

¹Excluding putative recombinant and laterally transferred alleles

²Mean of pairwise distances within clade

³Mean of pairwise distances between clades

⁴Mean of all pairwise distances

Brackets indicate significant differences between mean values based on all loci concatenated (independent t-test, P<0.01)

Phylogenetic differentiation among isolates

Bayesian inference and maximum likelihood analyses based on partial nucleotide sequences of each protein-coding locus differentiate the 25 isolates into two distinct and well supported clades (posterior probabilities = 1.0, bootstrap proportions ≥ 96%), hereafter designated clades I and II (Table 1; Figure 1). Tree topologies inferred by maximum likelihood analyses of amino acid sequences were consistent with those inferred from nucleotide sequences (Figure S3), with the exception of recA, for which only a single amino acid change in a single allele was observed. Mean interclade variation in nucleotide sequence is significantly greater (independent t-test, P≤0.01) than intraclade variation for all loci (Table 3). With five exceptions (recA in T8412, T8503 and T8513, celAB in PMS-535T.S.1b.3 and gyrB in T8510), clade assignments for each isolate were supported by analyses of all alleles of the five protein coding loci and by the concatenated sequence set (Table 1, Figure 1).

Figure 1. Phylogenetic trees inferred for T. turnerae isolates based on partial sequences of five protein-coding loci and the concatenated set.

The displayed trees were inferred using Bayesian inference and are identical in topology with respect to all well supported nodes (posterior probability ≥ 0.90, bootstrap proportions ≥ 70%) to trees inferred from the same data by Maximum Likelihood (data not shown). Numerical values at the nodes represent Bayesian posterior probability estimates ≥ 0.90 (left) and maximum likelihood bootstrap proportions ≥ 70% (right) determined for the corresponding node in a topologically identical Maximum Likelihood consensus tree. Trees are midpoint rooted. Circles, clade I; diamonds, clade II; squares, putative recombinant alleles.

Phylogenetic relationship of T. turnerae to other gamma proteobacteria

Phylogenetic analyses based on 16S rRNA, gyrB, rpoB and recA genes (Figure S2) indicate that all examined T. turnerae isolates form a monophyletic taxon most closely related to gamma proteobacteria of the order Alteromonadales.

Host specificity

All isolates examined, with the exception of T0609 and T0611, which were isolated from the same host specimen, represent unique genotypes containing at least one allele or combination of alleles not observed in any other isolate. In at least five host species (Dicyathifer manni, Lyrodus pedicellatus, Bankia gouldi, Bankia rochi, Teredo navalis, and Teredo clappi) distinct genotypes of T. turnerae were obtained from specimens collected in different locations. Two distinct genotypes of T. turnerae were obtained from a single specimen of L. pedicellatus (PMS-509L.S.1a.4 and PMS-509L.S.1a.6). Although genotypes differed, all specimens of each host species examined displayed symbiont clade specificity, i.e. each harbored T. turnerae isolates of the same clade. Symbiont clade specificity was observed among specimens of host species collected contemporaneously as well as those separated in time by up to 27 years.

Geographic distribution of collection sites

Figure 2 depicts the geographic distribution of collection sites for host specimens and associated T. turnerae isolates and clades.

Figure 2. Map depicting geographic distribution of collection sites for host specimens and associated T. turnerae isolates and clades.

Dashed lines indicate approximate northernmost (top) and southernmost (bottom) boundaries of occurrence of the Family Teredinidae. Circles, clade I; diamonds, clade II.

Distribution of T. turnerae clades among host taxa

Clade I isolates were found in host species representing the genera Dicyathifer, Neoteredo, Teredora, and Nausitora. Clade II isolates were found in host species representing the genera Teredo and Lyrodus. Isolates representing clade I and clade II were found in the genus Bankia, in the species B. gouldi (clade I), B. australis and B. rochi (clade II) (see Figure 3).

Figure 3. Occurrence of T. turnerae clades I and II mapped to host phylogeny.

Occurrences of symbiont clade I and clade II isolates are shown here mapped to a phylogram depicting the proposed evolutionary relationships among their respective hosts. The displayed phylogram is a subtree excerpted from a larger tree, originally published in [10], that includes 38 additional taxa. The displayed trees were inferred using Bayesian analysis of concatenated partial sequences of small and large subunit nuclear rRNA genes. Numerical values that appear at selected nodes indicate posterior probabilities ≥ 0.90 determined from Bayesian analyses (left) and bootstrap proportions ≥ 70% (right) determined for a topologically identical consensus tree inferred by Maximum Likelihood. Details of these analyses are found in [10]. Circles, clade I; diamonds, clade II.

Recombination between T. turnerae clades I and II

MaxChi, Chimaera, and 3Seq provided significant support (P≤0.01) for the conclusion that the celAB allele observed in isolate PMS-535T.S.1b.3 is the result of a recombination event between major parent T8602 (clade I) and minor parent T7902 (clade II) with breakpoints starting at position 372 of the aligned sequence set (Table 4). Similarly, results of analyses using GENECONV, MaxChi, BootScan, and 3Seq gave significant support (P≤0.01) for the conclusion that the gyrB allele observed in isolate T8510 results from a recombination event between major parent PMS-683H.S.1a.9 (clade II) and minor parent T8509 (clade I) with breakpoints starting at position 597 and ending at position 772 (Table 4). No difference in sequence length and no interruptions in reading frame were observed between the putative recombinant alleles and the consensus sequences of clades I and II. PCR amplification of the putative recombinants was reproducible in independent experiments using DNA extracts derived from the progeny of separate single colonies and so the putative recombinants are unlikely to be the result of PCR artifacts.

Table 4.

Evidence for proposed recombination events

	P value ≤ 0.01 scaled using a Bonferroni correctiona						Position of breakpointsb	Recombinant allele (major and minor parents)
Locus	RDP	GENECONV	BootScan	MaxChi	Chimaera	3Seq
celAB	-	-	-	6.884 × 10−4	5.935 × 10−3	1.122 × 10−4	372c	PMS-535T.S.1b.3 (T8602 and T7902)c
gyrB	-	2.427 × 10−4	3.861 × 10−8	7.166 × 10−3	-	1.161 × 10−7	597, 777c	T8510 (PMS-683H.S.1a.9 and T8509)c

Only P values ≤ 0.01 are displayed

^bIn the aligned dataset

^cInferred by 3Seq

Evidence of lateral gene transfer

Evidence of lateral gene transfer was observed in the recA loci of isolates T8412, T8503, and T8513; all isolated from host species collected in the western Atlantic (Table 1). For each of these isolates the PCR product obtained was ~1600 nucleotides in length, ~740 nucleotides longer than the clade I and clade II consensus sequences. The sequence of these amplicons revealed fragments of two tandemly repeated, contiguous, non-identical recA loci (S4a), hereafter referred to as recA₁ and recA₂. In each isolate, sequence determined for the upstream locus, recA₁, encodes 163 amino acid residues corresponding to, and displaying 99.8% nucleotide sequence identity with, the 5’ end of the T. turnerae clade II consensus sequence. This was followed by a short in frame sequence region encoding 35 amino acids without detectable nucleotide or amino acid homology to any sequence in the GenBank nonredundant database. This in turn was followed by a region of 22 in-frame codons including 3 stop codons and a second sequence encoding 288 amino acid residues corresponding to the 5’ end of a second putative recA coding sequence, recA₂, ending at the reverse primer. BLAST analysis of the observed fragment of the recA₂ locus against the Genbank non-redundant protein database showed this sequence to be most closely related (88.3% amino acid and 70.9 % nucleotide identity) to the recA gene of Aeromonas simiae (Genbank accession AEE38933.1). This sequence displayed 85.9% amino acid identity with the consensus sequences of the T. turnerae recA clades I and II. The two recA coding sequences also differ in %GC-content. The recA₁ sequences displayed 53%GC-content; closely matching the clade I and II consensus sequences (53%) as well as the %GC-content of the complete genome of T. turnerae isolate T7901 (50.8%) [50]. The recA₂ sequences displayed 41%GC-content. To exclude the possibility that this observation was the result of a PCR artifact, we independently amplified and sequenced fragments spanning the junction of the two recA loci using PCR primers specific to each of the detected recA alleles starting with genomic DNA from each of the three isolates, T8412, T8503, and T8513.

Analyses of selection

Analyses of synonymous and nonsynonimous substitutions, using four methods to determine selection, indicate an overall pattern of purifying selection for examined coding sequences (Table S2). No codon position received significant support for positive selection by more than one of the four methods applied.

Analysis of cophylogeny of hosts and symbionts

Randomization testing using TreeMap v.2 demonstrated that, given the observed data, neither the null hypothesis, that the symbiont phylogeny is no more congruent with the host phylogeny than would be a random symbiont tree with the same taxon set, nor the reciprocal null hypothesis, that the host phylogeny is no more congruent with the symbiont phylogeny than would be a random host tree with the same taxon set, could be rejected (P≤0.05).

Discussion

Here we report evidence that 25 isolates of the shipworm symbiont species T. turnerae, isolated from specimens representing a broad cross section of the phylogenetic and geographic range of their host species, form two distinct clades; a result that is well supported by independent analyses of each of five protein-coding loci examined and by the concatenated data set. The observed pattern of diversification for all loci demonstrates comparatively large sequence variation between clades as compared to low variability within clades (Table 3). Despite low within-clade variability, 24 of 25 examined isolates of this widespread shipworm symbiont species represent unique genotypes with respect to the examined loci.

The pattern of distribution of the two identified clades of T. turnerae among host species suggests a degree of specificity in sorting of symbiont clades among host species. Each host species examined was observed to be associated with a single symbiont clade. However, the reciprocal relationship was not observed. Instead, each symbiont clade is associated with several host species. Also, although each host species was associated with a single T. turnerae clade, distinct genotypes of T. turnerae were found within different specimens of the same host species, and in one case within a single specimen of one host species. More comprehensive sampling of isolates from individual host species and specimens will be required to evaluate the extent of variability that may occur among T. turnerae isolates within and among shipworm species.

The specific association of each host species with just one T. turnerae clade was observed regardless of when and where the host specimens were collected. For example, all specimens of Dicyathifer manni were found to be associated with T. turnerae isolates belonging to clade I, regardless of whether the host specimens were collected in Townsville, Australia in 1986 or in Bohol, Philippines in 2009 (Table 1). Similarly, clade II isolates were obtained from Bankia rochi whether collected in Karachi, Pakistan in 1984 or Sao Paulo, Brazil in 1985. Clade II isolates were obtained from specimens of L. pedicellatus collected in Cataban Island, Philippines in 2009 and in Long Beach, CA in 1979. Also, clade II isolates were obtained from descendants of the Long Beach specimens of L. pedicellatus even after maintenance in laboratory culture in natural seawater for twenty-seven years. These results suggest that the specific associations between individual host species and their respective T. turnerae clades are stable over long periods of time and great distances of geographic separation.

Consistent with these observations, host specimens and species harboring clade I and host specimens and species harboring clade II of T. turnerae were determined to be globally distributed (Figure 2) and to have vast and overlapping ranges. In half of the broadly distributed geographic locales examined, host specimens harboring clade I and host specimens harboring clade II isolates were found in close proximity, sometimes in the same piece of wood. These results suggest that geographic location may not be a critical determinant governing the occurrence of T. turnerae clades among host specimens and species. It should be noted, however, that although the sampling performed here was phylogenetically and geographically broad, it was not comprehensive. More comprehensive sampling will be required to establish the geographic dependence, or independence, of host-symbiont sorting in the T.turnerae-Teredinidae symbioses.

Similarly, the distribution or sorting of T. turnerae clades among host specimens examined here does not conform to host species boundaries as defined by anatomical [6] and/or molecular criteria [7]. On the contrary, symbiont clades are shared across broadly diverse host species (Table 1; Figure 3). Therefore, it seems unlikely that the observed divergence of T. turnerae into two distinct clades is explained by host and symbiont cospecieation. Cophylogeny at higher taxonomic levels is another possible explanation. When mapped to a previously published phylogenetic tree inferred for the family Teredinidae, clade I isolates appear to be primarily associated with more deeply branching taxa while clade II strains appeared in more recently emergent species [7] (Figure 3). However, this observation should be interpreted with caution. Not all of the host species sampled in this study were included in the previous phylogenetic analysis (and vice versa), sequences used in that study were obtained from different specimens than those sampled in this study, sampling across host taxa was not comprehensive, and a different set of genes was used to infer host and symbiont phylogeny. Moreover, randomization tests show that the observed congruence of host and symbiont phylogenies is not significantly greater than might be expected to occur by chance among random phylogenies with the same number of taxa. Thus, analysis of additional host taxa and symbiont isolates will be necessary to determine to what extent the observed differentiation among symbiont clades may reflect coevolution of hosts and symbionts. We note however that the host taxa are well differentiated by analysis of highly conserved ribosomal genes, indicating comparatively ancient divergence, while within clades the symbiont strains are not well differentiated by more rapidly evolving protein coding genes, indicating more recent diversification. Therefore, while examination of additional genes may improve resolution among isolates within clades, this in itself should not be expected to improve congruence between host and symbiont trees.

Given the strong support observed for differentiation of T. turnerae isolates into two distinct clades, it is reasonable to ask whether this differentiation is the result of genetic drift, selection, or both. Analysis of the ratio of nonsynonymous and synonymous mutations can be used to infer positive or negative selection when this value either exceeds or falls short of the random expectation. Such analyses suggest that the examined gene loci are largely under negative selection (as might be expected for core functional genes) and indicate that the majority of the phylogenetic signal distinguishing clades in nucleotide sequence analyses is due to synonymous substitutions. Nonetheless, amino acid differences are observed in all loci, and with the exception of recA, which contains just a single amino acid substitution, trees inferred on the basis of amino acid sequences support the same clade boundaries as do nucleotide sequence trees (Figure S3), indicating that non-synonymous substitution also contributes to phylogenetic signal. By the criteria applied here (P≥0.05 or REL Bayes Factor ≥100, posterior probability ≥0.95), no codon position received significant consensus support for positive selection from more than a single analytical method (Table S2). Based on these considerations it seems likely that drift rather than selection is responsible for the observed sequence differences among the examined loci. We note however that the examined genes, with the exception of celAB, are considered to be components of the core genome and so may not be strongly influenced by factors in the external environment. This does not preclude the possibility that other genes may be under positive selection.

While the majority of the 125 alleles examined in this study can be assigned unambiguously to either T. turnerae clade I or to clade II, exceptions were observed involving three loci (celAB, gyrB and recA) and five isolates (T8412, T8503, T8510, T8513, and PMS.535T.S.1b.3). These exceptions shed light on the extent to which members of the two T. turnerae clades are isolated genetically. Putative recombinant alleles of the celAB and gyrB loci, apparently derived from recombination between clade I and clade II parent alleles, were observed in two T. turnerae isolates. Several analytical methods provide significant support (P≤0.01; Table 4) for these recombination events, suggesting that genetic exchange has occurred between members of the two clades since their divergence. This observation is not unprecedented. Evidence for symbiont exchange among host species [51] and homologous recombination [52] among symbiont isolates has been reported for gill endosymbionts of vesicomyid clams, indicating that acquisition of the intracellular symbiotic habit may not preclude genetic exchange, even among symbionts of host species presumed to harbor nearly monospecific symbiont populations and to acquire their symbionts primarily via vertical transfer. While the mode of transmission of T. turnerae in the shipworm symbiosis is unknown, these results suggest that some strains of the two T. turnerae clades have, at some point in time, been in sufficiently close physical proximity to allow genetic exchange. This might indicate that T. turnerae is environmentally transferred and that opportunity for genetic exchange has occurred in the external environment. Alternatively, concurrent multiple infections of individual hosts by members of both clades may occur with sufficient frequency to provide opportunities for genetic exchange between clades within the host tissues.

Similarly, features of the recA loci of three T. turnerae isolates (T8412, T8503 and T8513) suggest that genetic exchange between these and other bacterial species is possible. In these isolates portions of two distinct recA loci were observed, one of which, recA₁, appears to be a native locus that has been truncated by insertion of a sequence containing a second, apparently exogenous, recA locus (recA₂).

Several types of evidence support the exogenous origin of recA₂. First, with few reported exceptions, e.g. [53, 54], recA is present as a single copy gene in bacterial genomes examined to date [55], including the closed and finished genome sequence of T. turnerae isolate (T7901)[50]. Consistent with this, a single recA sequence was detected by PCR in genomes of all other T. turnerae isolates examined here. The truncated sequence of the recA₁ allele in each of the three isolates closely matches the T. turnerae clade II consensus sequence. All other loci examined in these isolates were also consistent with T. turnerae clade II. In contrast, in each of these isolates the sequence of the recA₂ allele is significantly divergent from the recA consensus sequences of both T. turnerae clades I and II, and shows equal or greater identity to recA sequences of distantly related gammaproteobacteria (Figure S4). The recA₂ sequences also differ significantly in %GC-content (41%) as compared to T. turnerae recA₁ sequences (53%) and to the complete genome of T. turnerae isolate T7901 (50.8%)[50]. Taken together, these observations suggest that the recA₂ locus in these three isolates entered these genomes as the result of a horizontal gene transfer event that disrupted the endogenous recA locus. It is most parsimonious to conclude that this event occurred after the divergence of clades I and II. Again this suggests that T. turnerae, regardless of its status as intracellular endosymbiont, is not entirely isolated from genetic exchange with other bacterial species. It remains unclear whether this putative genetic exchange was facilitated by exposure to other bacterial species in the external environment or to exchange with other symbiont species that coexist with T. turnerae within the host tissue.

In summary, sequence analysis of T. turnerae isolates examined here indicates the emergence of two distinct lineages within this ubiquitous shipworm symbiont species, suggesting diversification in the presence of a persistent barrier to genetic exchange between but not among members of the two clades. While each host species examined appears to associate with just one of the two T. turnerae clades, each T. turnerae clade is associated with several host species, each having wide and overlapping geographic distributions. To this extent, the observed pattern of diversification of T. turnerae does not appear to be consistent with, and therefore explained by, host geography or host speciation. However, additional sampling will be required to determine the extent to which geography, environment or host and symbiont cophylogeny contribute to the assortment of T. turnerae clades among teredinid hosts. The results presented here also indicate that, in spite of the apparent genetic isolation of the two clades, both recombination between clades and acquisition of foreign DNA may contribute to the diversification of T. turnerae genomes. Finally, based on the results observed here, it is interesting to speculate that divergence among subpopulations of established endosymbiont types, rather than, or in addition to, de novo acquisition of new bacterial types from the environment, may have contributed to the observed coexistence of multiple endosymbiont types in shipworms.