Strict cospeciation of devescovinid flagellates and Bacteroidales ectosymbionts in the gut of dry-wood termites (Kalotermitidae)


E-mail; Tel. (+49) 6421 178 701; Fax (+49) 6421 178 999.


The surface of many termite gut flagellates is colonized with a dense layer of bacteria, yet little is known about the evolutionary relationships of such ectosymbionts and their hosts. Here we investigated the molecular phylogenies of devescovinid flagellates (Devescovina spp.) and their symbionts from a wide range of dry-wood termites (Kalotermitidae). From species-pure flagellate suspensions isolated with micropipettes, we obtained SSU rRNA gene sequences of symbionts and host. Phylogenetic analysis showed that the Devescovina spp. present in many species of Kalotermitidae form a monophyletic group, which includes also the unique devescovinid flagellate Caduceia versatilis. All members of this group were consistently associated with a distinct lineage of Bacteroidales, whose location on the cell surface was confirmed by fluorescence in situ hybridization. The well-supported congruence of the phylogenies of devescovinids and their ectosymbionts documents a strict cospeciation. In contrast, the endosymbionts of the same flagellates (‘Endomicrobia’) were clearly polyphyletic and must have been acquired independently by horizontal transfer from other flagellate lineages. Also the Bacteroidales ectosymbionts of Oxymonas flagellates present in several Kalotermitidae belonged to several distantly related lines of descent, underscoring the general perception that the evolutionary history of flagellate–bacteria symbioses in the termite gut is complex.


Anaerobic flagellate protozoa play a critical role in lignocellulose digestion in the hindgut of phylogenetically lower termites (Cleveland, 1923; Hungate, 1955; Breznak and Brune, 1994). The flagellates are associated with a diverse community of ectosymbiotic and endosymbiotic bacteria, which represent several phyla and are typically specific for their respective hosts (for reviews, see Brune and Stingl, 2005; Brune, 2006; Ohkuma, 2008). Neither the flagellates nor the bacterial symbionts are in permanent culture, but phylogenetic analysis of their SSU rRNA genes provided first hints that at least some of these symbiotic associations might have evolved by cospeciation (for review, see Ohkuma, 2008).

Cospeciation, resulting from intimate and long-standing association of host and symbiont, exists if molecular phylogenies of hosts and symbionts are significantly more similar than would be expected due to chance alone, and has been observed in various types of symbioses (e.g. Baumann et al., 1997; Peek et al., 1998; Hosokawa et al., 2006; Hughes et al., 2007). To date, two examples of cospeciation between termite gut flagellates and their bacterial endosymbionts have been described, involving Pseudotrichonympha spp. and Bacteroidales (Noda et al., 2007), and Trichonympha spp. and ‘Endomicrobia’ (Ikeda-Ohtsubo and Brune, 2009).

Endosymbionts located within the cytoplasm of a flagellate are easily transferred to the daughter cells when the host divides (Ikeda-Ohtsubo and Brune, 2009). Such a vertical transmission is less likely in the case of ectosymbionts. If the cells are not firmly attached to the host's surface, they may dissociate, leading to a permanent loss, and reattach to another host, resulting in a horizontal transfer of symbionts between host lineages, as was shown for ectosymbiotic Spirochaetes (Noda et al., 2003). Although such mechanisms would reduce the chances of cospeciation in ectosymbiosis, the close phylogenetic relatedness of Bacteroidales ectosymbionts of three devescovinid flagellates (Devescovina spp. and Caduceia versatilis) suggests that also ectosymbionts of termite gut flagellates can be vertically transmitted (Hongoh et al., 2007).

Devescovinid flagellates are present mainly in the gut of dry-wood termites; they have been classified on the basis of their morphological characteristics and the presence or absence of ectosymbiotic bacteria (Kirby, 1941; 1942; 1945). The highest number of species has been assigned to the genus Devescovina (Kirby, 1941), yet knowledge about their phylogeny is scarce (Gerbod et al., 2002). Each of the 20 Devescovina species carry ectosymbionts (Kirby, 1941; Branke, 1996; Radek et al., 1996; Noda et al., 2006a); these laterally attached filamentous rods may cover the entire surface of the host cell (Fig. 1A and B). The consistent presence of such rods on the surface of all Devescovina species offers an excellent opportunity for studying cospeciation of ectosymbionts with their flagellate host.

Figure 1.

Scanning electron micrographs of Devescovina sp. (A) and Oxymonas sp. (C) and their ectosymbiotic bacteria (B, D) in Neotermes castaneus (batch BAM-2), exemplifying the two major groups of flagellates investigated in this study.
A. Devescovina sp. has a spindle-shaped body, three anterior flagella (fla), and a thicker, ribbon-shaped, recurrent flagellum (flr) typical of this genus (Kirby, 1941).
B. As in other Devescovina spp. (Branke, 1996; Radek et al., 1996; Noda et al., 2006a), the entire flagellate is covered by a layer of uniform, filamentous bacteria; the arrows indicate the two ends of a single cell.
C. Oxymonas sp. has a rostellum (ro), typical of this genus (Cleveland, 1950; Brugerolle and König, 1997); as in another Oxymonas sp. (Noda et al., 2006b), the flagellate is colonized by two morphotypes of ectosymbiotic bacteria (D): long, slender rods with tapered ends (white arrows), and short, thick rods with rounded ends (black arrows), sometimes in division (arrowhead). Scale bars: 20 μm (A, C), 3 μm (B, D).

In the present study, we investigated the molecular phylogenies of Devescovina spp. and their bacterial symbionts. Species-pure suspensions of flagellates from numerous species of dry-wood termites (Table 1) were isolated with micropipettes. The phylogenies of host flagellates and bacterial symbionts were reconstructed and compared based on their SSU rRNA gene sequences. Fluorescence in situ hybridization was carried out to determine the exact location of Bacteroidales. Since Bacteroidales ectosymbionts also occur on Oxymonas sp. in Neotermes koshunensis (Noda et al., 2006b), the ectosymbionts of the Oxymonas spp. (exemplified by Fig. 1C and D) present in the termites used in this study were included in the phylogenetic analysis. Finally, we compared the phylogenies of the devescovinid flagellates with that of their host termites using the cytochrome oxidase II (COII) gene of the termite as a phylogenetic marker.

Table 1.  Phylotypes of flagellates and their bacterial symbionts in the clone libraries obtained from capillary-picked suspensions of devescovinid and oxymonadid flagellates from different termites.
TermitesaFlagellate suspensionsPhylotypes in clone libraries (abundance)b
  • a. 

    Country of origin and GenBank accession number of COII gene sequence.

  • b. 

    Number of clones per number of total clones tested. Additional minor phylotypes in the 16S rRNA gene clone libraries belonged to Alphaproteobacteria, Mycoplasmatales, Spirochaetes and Verrucomicrobia; they are not specified in the table but were submitted to GenBank (accession numbers: FN377777FN377800).

  • c. 

    Sequence is virtually identical to that of phylotype AM747389 (Ikeda-Ohtsubo et al., 2007).

  • d. 

    Sequences have been already reported in the context of other studies (Pester and Brune, 2006; Ikeda-Ohtsubo et al., 2007).

  • e. 

    Kirby (1941) described Devescovina sp. in C. longicollis as D. lepida. However, the phylotype obtained from C. longicollis is clearly different from D. lepida in N. castaneus (see Fig. 2).

  • f. 

    Clone library was prepared with ‘Endomicrobia’-specific primers (Stingl et al., 2005).

  • g. 

    PCR of the SSU rRNA gene of the flagellate was not successful with the DNA preparation used for the 16S rRNA gene clone library. Unfortunately, neither of two replacement batches of C. secundus contained Oxymonas flagellates, even though the termites were collected at the same time and place.

  • h. 

    The clone library did not contain any ‘Endomicrobia’ sequences.

  • i. 

    These sequences are nearly identical to the previously reported sequences of C. brevis and C. dudleyi collected in Australia (Thompson et al., 2000).

  • j. 

    The sequences of C. versatilis and its ectosymbiont are from previous studies (Noël et al., 2007; Hongoh et al., 2007); the abbreviation CcCv was introduced for the purpose of Fig. 5. The cytochrome oxidase II gene sequence of C. cavifrons was determined with termites collected in Florida, USA, and identified by R. Scheffrahn. Light microscopy of the gut homogenate showed the unique morphotype of C. versatilis (‘Rubberneckia’).

  • k. 

    The presence of ‘Endomicrobia’ was not reported (Hongoh et al., 2007).

Neotermes castaneus
Cuba FN377805
Devescovina lepidaNcDvL11 (4/8) FN377759cNcDvLB13 (10/24) FN377751NcDv-1 (3/24) AB298058d
NcDvL20 (4/8) FN377760NcDvLB23 (9/24) FN377752
Devescovina artaNcDvA25 (10/10) FN377761NcDvAB05 (9/10) FN377753NcDvAE10 (1/10) FN377767
Oxymonas sp.NcOxA (3/3) AB326383dNcOxAB12 (8/26) FN377773NcOx-1 (2/26) AB298081d
NcOxAB26 (6/26) FN377774
Cryptotermes longicollis
Mexico FN377806
Devescovina sp.eClDv14 (4/4) FN377762ClDvB13 (12/18) FN377754ClDvE07 (4/18) FN377768
Cryptotermes secundus
Australia DQ278261d
Devescovina sp.CsDv11 (8/8) FN377763CsDvB04 (9/9) FN377755CsDvE04 (3/3)f FN377769
Oxymonas sp.Not determinedgCsOxB27 (10/27) FN377775h
CsOxB40 (4/27) FN377776
Cryptotermes brevisi
Brazil FN377807
Devescovina sp.CbDv02 (7/7) FN377764CbDvB11 (22/30) FN377756CbDvE07 (1/30) FN377770
Cryptotermes dudleyii
Kenya FN377808
Devescovina sp.CdDv09 (4/4) FN377765CdDvB38 (7/10) FN377757CdDvE17 (2/10) FN377771
Cryptotermes havilandi
Ghana FN377809
Devescovina sp.ChDv22 (5/5) FN377766ChDvB35 (16/20) FN377758ChDvE11 (2/20) FN377772
Cryptotermes cavifronsj
USA FN377810
Caduceia versatilisCcCv DQ855405CcCv-03 AB299517k


Morphology of Devescovina spp. and their ectosymbionts

Almost all dry-wood termites (Kalotermitidae) investigated in this study harboured only one morphotype of Devescovina flagellates (Table 1). An exception was Neotermes castaneus, which contains two species, the larger Devescovina lepida and the smaller Devescovina arta (Kirby, 1941). In batch BAM-1, used for molecular analyses, there was a distinctly bimodal width distribution of the Devescovina cells (Fig. S1), which allowed the two morphotypes to be picked separately (see below).

Scanning electron microscopy was conducted with a different batch of N. castaneus (BAM-2; Fig. 1A and B). Although N. castaneus individuals of both batches had identical COII genes, batch BAM-2 harboured only a single morphotype of Devescovina (morphological details are summarized in Table S1). Unfortunately, it was not possible to identify the phylotype of the flagellate in batch BAM-2 because at this stage no specimens were available for molecular analyses.

Phase-contrast light microscopy showed the presence of filamentous rods on the surface of all Devescovina spp. investigated, which is in agreement with the original genus description of Kirby (1941). The Devescovina species in Cryptotermes secundus was also associated with spirochetes, which were typically attached at the posterior end of the flagellate.

Phylogenetic analysis of devescovinid flagellates

Flagellate suspensions prepared by capillary picking of cells with the same morphotype each yielded only a single phylotype of SSU rRNA genes (> 99.8% sequence similarity), corroborating the presence of only one species in each suspension. An exception was the flagellate suspension of D. lepida, which contained two highly similar variants (0.2% sequence divergence), whose existence was confirmed by their presence also in a second clone library prepared from a different PCR product and by signature analysis of the resulting sequences (not shown).

Phylogenetic analysis revealed that all phylotypes form a monophyletic cluster, which included the sequences of Caduceia versatilis (Noël et al., 2007), Devescovina sp. NK9 (Gerbod et al., 2002), and several other previously published flagellate sequences from termite gut homogenates that had been tentatively assigned to the genus Devescovina (Fig. 2). To increase the robustness of the tree topology, we also included the sequences of several other devescovinid flagellates obtained from whole gut homogenates of the termites used in this study (details not shown).

Figure 2.

Phylogenetic tree of devescovinid flagellates from dry-wood termites (host termite in parentheses). Sequences from this study are shown in bold. Tree topology is based on maximum-likelihood analysis of SSU rRNA gene sequences (1316 unambiguously aligned nucleotide positions) and was verified by neighbour-joining and maximum parsimony analyses. Additional node support was obtained by bootstrap analysis (DNAPARS, 100 replicates; ●, > 70%; ○, > 50%). Vertical lines denote monophyletic groups of flagellates representing different genera of devescovinids. Previously unidentified sequences from gut homogenates (shown with asterisks) were assigned to their respective genera based on their phylogenetic position. The host termite of Devescovina sp. D16 and parabasalid clone NJ1, given as Neotermes sp. in the PhD thesis of Branke (1996) and as Neotermes jouteli in the respective GenBank entries (accession numbers: X97974 and X97975), was most probably Neotermes castaneus; N. jouteli does not harbour Devescovina sp. (Kirby, 1942; our study). The specimens of N. jouteli used in our study (COII gene sequence accession number FN377811) were identified by R. Scheffrahn.

Bacteroidales ectosymbionts of Devescovina

The 16S rRNA gene clone libraries of the same flagellate suspensions were each dominated (more than 90% of the clones) by a single phylotype (> 99.8% sequence similarity) of Bacteroidales (Fig. 3). The only exception was again the flagellate suspension of D. lepida, which not only contained two closely related phylotypes of flagellates (see above) but also yielded two phylotypes of Bacteroidales in roughly equal proportion (0.6% sequence divergence). Again, the existence of two variants was confirmed by two independent clone libraries from different PCR products and signature analysis of the resulting sequences (not shown).

Figure 3.

Phylogenetic tree illustrating the relationships of ‘Armantifilum devescovinae’ and other Bacteroidales obtained from termite guts; sequences from this study are shown in bold. Names of termite species are given in parentheses. Tree topology is based on maximum-likelihood analysis of 16S rRNA gene sequences (1273 unambiguously aligned nucleotide positions), and was verified by neighbour-joining and maximum parsimony analyses. Additional node support was obtained by bootstrap analysis (DNAPARS, 100 replicates; ●, > 90%; ○, > 70%). Vertical lines denote termite gut clusters of Bacteroidales based on Ohkuma and colleagues (2002); not all available sequences (including the basal cluster I) are shown. The tree was rooted with representatives of other phyla.

Phylogenetic analysis showed that all Bacteroidales sequences obtained from the Devescovina suspensions clustered together (Fig. 3), including also two Bacteroidales clones previously obtained from Devescovina spp. of other dry-wood termites (Noda et al., 2006a). Together with the ectosymbiont of C. versatilis (Hongoh et al., 2007), they formed a monophyletic group within termite gut Bacteroidales Cluster V (Ohkuma et al., 2002).

To determine the exact location of the Bacteroidales obtained from the flagellate suspensions, we carried out in situ hybridization with the oligonucleotide probe CF319a (Manz et al., 1996) specific for most Bacteroidales. For all Devescovina spp. investigated in this study, the probe hybridized with filamentous bacteria located on the surface of the flagellate cells. In each of the termites investigated, all Devescovina cells present in the hindgut fluid were consistently colonized by such filamentous ectosymbionts. Counter-staining with the general bacteria probe EUB338 revealed that in all Devescovina spp. – with the exception of spirochetal forms in the Devescovina sp. of C. secundus– any bacteria belonging to other phyla, such as ‘Endomicrobia’, were apparently located within the flagellate cell.

To investigate the specificity of the ectosymbionts for their Devescovina host, we designed the group-specific probe DVB178, which exactly matches sequences of the closely related Bacteroidales ectosymbionts of D. lepida and D. arta in N. castaneus. Probe DVB178 always co-hybridized with the filamentous ectosymbionts of the two Devescovina spp. that hybridized with probe CF319a (Fig. 4), but never with bacterial symbionts of other flagellates or with bacteria in the gut fluid.

Figure 4.

Fluorescence in situ hybridization of the hindgut content of Neotermes castaneus with an oligonucleotide probe (DVB178) specific for the Bacteroidales ectosymbionts (‘Armantifilum devescovinae’) associated with the two Devescovina spp. from this termite. The probe hybridized exclusively with the filamentous ectosymbionts of the two species. To illustrate the location of the bacteria, the images were taken with different focal planes.
A. Devescovina arta, focus at centre (arrows indicate bacterial layer on flagellate surface, arrowheads indicate autofluorescence of residual wood particles).
B. Devescovina lepida, focus on surface. Scale bars, 10 μm.

Bacteroidales ectosymbionts of Devescovina spp., also including the previously investigated phylotypes (Noda et al., 2006a), were named ‘Candidatus Armantifilum devescovinae’ (see Discussion; Fig. 3).

Cospeciation analysis

We tested for cospeciation of the seven symbiotic pairs of Devescovina flagellates and Bacteroidales ectosymbionts (‘A. devescovinae’) obtained from the flagellate suspensions. The previously reported symbiotic pair of C. versatilis and Bacteroidales clone CcCv-03 (Hongoh et al., 2007; Noël et al., 2007), which clustered with the phylotypes obtained in this study, was included in the analysis. Although also the Devescovina clones Cd3 and Nk2 and the Bacteroidales clones CdD3-1 and NkD2-1 (Ohkuma et al., 2000; Noda et al., 2006a) were good candidates for symbiotic pairs (see Figs 2 and 3), they were excluded from the analysis. Inclusion of these sequences, which were 90–120 bp shorter than those of our study, led to a loss of node support in the phylogenetic trees. Moreover, in the case of N. koshunensis, the exact pairing of host and symbiont phylotypes remains to be established (two host sequences and only one symbiont sequence were reported).

Detailed phylogenetic analyses showed that the trees of hosts and symbionts perfectly mirrored each other (Fig. 5). Reconciliation analysis (Treemap) produced seven cospeciation events. The linear relationship between the coalescence times for host and symbiont of each symbiotic pair, estimated from the maximum-likelihood (ML) distances, corroborated strict cospeciation of the two lineages (Fig. 6). In contrast, the phylogeny of the termites was largely incongruent with the phylogeny of their flagellates, documenting that cospeciation did not extend to the host termites of the symbiotic pairs (Fig. 7).

Figure 5.

Phylogenetic tanglegram of devescovinid flagellates and ‘Armantifilum devescovinae’. Tree topologies were estimated by maximum-likelihood (ML) analyses of SSU rRNA genes of both taxa. Node support labels (all in percentage, separated by slashes) indicate results of Bayesian posterior probability and bootstrap analysis (DNAPARS, 1000 replications; ML, 100 replications). The host tree is based on SSU rRNA gene sequences (1408 unambiguously aligned nucleotide positions) and was rooted with Trichomonas vaginalis (AY338474) and Tritrichomonas foetus (AY055799). The symbiont tree is based on 16S rRNA gene sequences (1363 unambiguously aligned nucleotide positions) and was rooted with Bacteroides fragilis (M11656) and Tannerella forsythensis (X73962). Filled circles mark cospeciation events inferred from reconciliation analysis in Treemap (< 0.001); because of the unresolved pairing of the two phylotypes of Devescovina lepida and their Bacteroidales ectosymbionts in Neotermes castaneus, only one phylotype each was considered. Dashed lines connect the symbiotic pairs and also indicate the termite harbouring the respective pair; for full species names of termites and flagellates and sequence accession numbers, see Table 1.

Figure 6.

Coalescence analysis of the phylogenies of host (devescovinids) and symbiont (‘Armantifilum devescovinae’). Coalescence times are the maximum-likelihood distances of the individual cospeciation events from the basal node (Fig. 5). The regression line has a slope of 0.51 (r2 = 0.98).

Figure 7.

Phylogenetic tanglegram of termites and their symbiotic devescovinid flagellates. The topology of the termite tree was estimated by maximum-likelihood analysis of the cytochrome oxidase II genes (182 unambiguously aligned deduced amino acid sites); the SSU rRNA gene tree of the flagellates is the same as in Fig. 5. Dashed lines connect the symbiotic pairs. Filled circles denote nodes with high bootstrap support (> 90% for the termite tree; for the flagellate tree, see Fig. 5). The termite tree was rooted with Kalotermes spp. (GenBank accession numbers: AF189100 and AF189102). For species names of flagellates and sequence accession numbers, see Table 1.

Endosymbionts of Devescovina

All Devescovina spp. investigated in this study contained endosymbionts affiliated with the ‘Endomicrobia’. They represented less than 10% of the clones in the bacterial clone libraries of the different Devescovina suspensions; in the case of the Devescovina sp. in C. secundus, they could be detected only with ‘Endomicrobia’-specific primers. The clones from each suspension formed unique phylotypes (> 99.8% sequence similarity) but fell into different clusters of ‘Endomicrobia’, comprising sequences previously obtained from flagellates of other phylogenetic groups (Fig. 8). The ‘Endomicrobia’ in one of the Devescovina sp. had been previously localized in situ with a phylotype-specific probe (Ikeda-Ohtsubo et al., 2007).

Figure 8.

Phylogenetic tree showing the relationships of ‘Endomicrobia’ symbionts of termite gut flagellates, based on 16S rRNA gene sequences; sequences obtained in this study are shown in bold. The tree is based on 1312 valid columns and was verified by neighbour-joining and maximum parsimony analyses. Additional node support was obtained by bootstrap analysis (DNAPARS, 100 replicates; ●, > 90%; ○, > 70%). The tree was rooted with representatives of other phyla. The names of termite species are in parentheses. Vertical lines indicate the ‘Endomicrobia’ sequences from dry-wood termites.

Bacteroidales ectosymbionts of Oxymonas

During in situ hybridization of the ectosymbionts of devescovinids, we noticed that also the Oxymonas spp. present in several of the dry-wood termites were covered with bacteria that hybridized with the oligonucleotide probe CF319a specific for most Bacteroidales. In all Oxymonas species, colonization was consistent but often patchy. Noda and colleagues (2006b) had already reported two types of Bacteroidales associated with an Oxymonas sp. from N. koshunensis, which were reportedly associated with the flagellate surface, but electron micrographs were not shown.

Scanning electron microscopy conducted with the Oxymonas sp. in N. castaneus BAM-2 showed that the surface of this flagellate is covered with two morphotypes of rod-shaped bacteria attached in an irregular pattern (Fig. 1C and D). Some specimens were also associated with spirochetes, although their number varied among individual flagellates. In those cells in which spirochetes were abundant, the density of rod-shaped ectosymbionts was markedly reduced. The morphological details are summarized in Table S1.

We obtained clone libraries of the 16S rRNA genes associated with capillary-picked Oxymonas suspensions from N. castaneus (BAM-1) and C. secundus. Each library contained two phylotypes of Bacteroidales, which together formed more than half of the clones in each library (Table 1), and several other, less abundant phylotypes (details not shown) that fell into Verrucomicrobia and Spirochaetes (GenBank accession numbers: FN377777FN377791). Phylogenetic analysis showed that the four Bacteroidales phylotypes, also including the two Bacteroidales ectosymbionts of Oxymonas sp. from N. koshunensis, are clearly polyphyletic, each forming distinct lineages in different, previously defined clusters of termite gut-associated Bacteroidales (Ohkuma et al., 2002; Fig. 3).


This is the first report of cospeciation of an ectosymbiotic bacterium with its protistan host. The molecular phylogenies of ‘Candidatus Armantifilum devescovinae’ (Bacteroidales) and devescovinid flagellates (i.e. Devescovina spp. and C. versatilis) perfectly mirror each other. While these ectosymbionts are highly specific to the host flagellates, there are strong indications that the endosymbionts (‘Endomicrobia’) of Devescovina spp. were acquired by multiple horizontal transfers from other gut flagellates. The strict cospeciation of devescovinids and ectosymbionts is maintained despite the horizontal transfer of flagellates among several species of dry-wood termites (Kalotermitidae), which underlines that cospeciation is not merely caused by a spatial separation of flagellates in their termite hosts, but is the result of an obligate symbiosis, leading to strict vertical transmission of ‘A. devescovinae’ by their flagellate hosts.

The results obtained by SSU rRNA gene analysis of capillary-picked Devescovina flagellate suspensions, followed by in situ localization of the sequences in hindgut contents, convincingly show that the filamentous ectosymbionts of all investigated Devescovina spp. form a monophyletic group among the Bacteroidales. Strict cospeciation of ‘A. devescovinae’ with their devescovinid hosts is documented by the fully supported congruence of their phylogenetic trees, and further corroborated by the highly significant results of cospeciation test and coalescence analysis (Hughes et al., 2007). Regression analysis of coalescence times suggests that the SSU rRNA gene of the symbionts evolved two times faster than that of the hosts (Fig. 6); an intercept close to zero indicates almost synchronous cospeciation (Hughes et al., 2007).

The Bacteroidales ectosymbionts previously obtained from two other Devescovina spp. (Cd3, Nk2; Noda et al., 2006a) also fall into the radiation of ‘A. devescovinae’, and the phylogenetic position of their hypothetical host flagellates (Cd3–CdD3-1 and Nk2–NkD2-1) indicates that they are cospeciating just like the symbiotic pairs included in this study. Evidently, ‘A. devescovinae’ represents the filamentous rods described already by Kirby (1941) as characteristic of the genus Devescovina. This is further supported by the fact that also the filamentous ectosymbiont of C. versatilis (Tamm, 1982; Hongoh et al., 2007), a flagellate that clusters within the Devescovina group (Noël et al., 2007; Fig. 2), is affiliated with the ‘A. devescovinae’ cluster. Since filamentous rods are absent on other devescovinid flagellates (Kirby, 1941; 1942; 1945; Strassert et al., 2009), it is safe to assume that the acquisition of the ancestor of ‘A. devescovinae’, a representative of the Bacteroidales Cluster V specific for termite guts, by the common ancestor of Devescovina species (and C. versatilis) gave rise to the extant symbiotic pairs by cospeciation.

The fact that C. versatilis and its filamentous ectosymbiont are one of these pairs corroborates the finding that the devescovinid flagellate C. versatilis (‘Rubberneckia’) clusters with species of the Devescovina group (Noël et al., 2007; Fig. 2). This raises the question whether C. versatilis belongs to the genus Caduceia (Kirby, 1942) or is a species of Devescovina. As morphological features that allow a separation of Devescovina species and Caduceia species (Duboscq and Grassé, 1927; Kirby, 1941; 1942) are weak, Kirby used the presence of filamentous ectosymbiotic rods as an important characteristic separating species in the genus Devescovina from those in all other devescovinid genera (including Caduceia). Notably, the devescovinid flagellate ‘Rubberneckia’ discovered by Tamm and Tamm (1974) was later described as a species of Caduceia (versatilis) by D'Ambrosio and colleagues (1999) even though it possesses filamentous rods similar to those present on Devescovina species (Tamm, 1982). The phylogenetic position of other Caduceia species must be studied in order to resolve this issue.

The flagellate communities in the gut of dry-wood termites are remarkably similar at the generic level (Kitade, 2004). Nevertheless, a single species of dry-wood termites may simultaneously harbour devescovinid flagellates of several genera, and similar morphotypes occur in different termites (Kirby, 1941; 1942, 1945; Yamin, 1979). This can be explained by a horizontal transfer of devescovinids among the termite species – a hypothesis that is supported also by the complete incongruence between the phylogenies of devescovinid flagellates and their termite hosts (Fig. 7). The fact that ‘A. devescovinae’ strictly cospeciated with its host flagellate, even in those cases where the presence of several devescovinid species in the same termite gut should have provided ample opportunities for an exchange of ectosymbionts, underscores the high specificity of this obligate symbiosis.

To date, only two instances of cospeciation in flagellate–bacterium symbioses have been documented; both occur in the termite gut. One is the symbiosis of Pseudotrichonympha flagellates with ‘Azobacteroides pseudotrichonymphae’ (Noda et al., 2007), the other is the symbiosis of Trichonympha flagellates with ‘Endomicrobium trichonymphae’ (Ikeda-Ohtsubo and Brune, 2009). Since Pseudotrichonympha spp. are cospeciating also with their host termites (Rhinotermitidae; Noda et al., 2007), it cannot be excluded that cospeciation with their bacterial symbionts is merely due to a spatial separation of the flagellate hosts within their respective termite species. However, in the case of Trichonympha spp., which show a horizontal transfer of flagellates among termite species of different families (Ikeda-Ohtsubo and Brune, 2009), strict vertical transmission of the symbionts within their host lineage is a prerequisite for explaining cospeciation of the symbiotic pairs. This is exactly the situation we observed here with Devescovina flagellates and ‘A. devescovinae’ in dry-wood termites.

Cospeciation of bacterial endosymbionts and their hosts is a widespread phenomenon (Clark et al., 2000; Sauer et al., 2000; Thao et al., 2000). However, the only ectosymbiotic bacterium reported to cospeciate with its host is ‘Candidatus Ishikawaella capsulata’ (Gammaproteobacteria), which attaches to the intestinal epithelia of stinkbugs (Hosokawa et al., 2006). Specific attachment sites for ectosymbionts, as reported for the flagellates Mixotricha paradoxa (Cleveland and Grimstone, 1964), Joenia annectens (Radek et al., 1996) and Staurojoenina sp. (Stingl et al., 2004), are absent in Devescovina species (Radek et al., 1996).

The driving force behind such intimate symbioses could be metabolic interactions (for a review, see Brune and Stingl, 2005). The most reasonable assumption is that the bacterial ectosymbionts serve as a source of essential amino acids or cofactors for their flagellate hosts. In Devescovina spp., phagocytosis of ectosymbionts has been evidenced by transmission electron microscopy (Radek et al., 1996; Noda et al., 2006a), indicating that the host flagellate uses ‘A. devescovinae’ as nutrient source. A more specific role in nitrogen metabolism would be in amino acid upgrading, as suggested for ‘E. trichonymphae’, the endosymbiont of the distantly related genus Trichonympha (Hongoh et al., 2008a). Although nitrogen fixation is generally absent in members of the phylum Bacteroidetes, the finding of a nifH gene and its expression in ‘A. pseudotrichonymphae’, the endosymbiont of the distantly related genus Pseudotrichonympha, documented the first case of nitrogen fixation by a symbiont of termite gut flagellate (Hongoh et al., 2008b). Other possible functions of ‘A. devescovinae’ include maintenance of the cytoskeletal structure of their host flagellate (Radek et al., 1996); consumption of hydrogen produced by the flagellates, as shown for ‘A. pseudotrichonymphae’ (Inoue et al., 2007); and protection of the oxygen-sensitive host from oxygen diffusing into the gut (Brune, 1998), as suggested for the Desulfovibrio symbionts of Trichonympha spp. (Sato et al., 2009).

Although strict cospeciation may also occur in other instances, it is not necessarily a rule governing the numerous flagellate–bacteria associations in termite guts. This is illustrated by the polyphyletic nature of the Bacteroidales sequences recovered from the capillary-picked Oxymonas spp., which fall, together with the Bacteroidales ectosymbionts of Oxymonas sp. from Neotermes koshunensis (Noda et al., 2006b), into several termite-gut clusters (Ohkuma et al., 2002; Fig. 3), indicating that Oxymonas flagellates must have acquired their ectosymbionts multiple times from a pool of free-living Bacteroidales. Likewise, the endosymbionts (‘Endomicrobia’) of Devescovina spp. must have been acquired independently by their respective host flagellates. As the closest relatives of the ‘Endomicrobia’ associated with Devescovina spp. are endosymbionts of phylogenetically unrelated flagellates from dry-wood termites, many parabasalid flagellates may have acquired their ‘Endomicrobia’ by horizontal transfer from other flagellates present in the same gut (Ikeda-Ohtsubo et al., 2007; Ohkuma et al., 2007). Given multiple acquisitions and horizontal transfer of other symbionts, the reasons for the strict cospeciation of ‘A. devescovinae’ and its devescovinid host remain enigmatic.

Description of ‘Candidatus Armantifilum devescovinae’′ lum. L. part. adj. armans arming; L. neut. n. filum filament; N.L. neut. n. armantifilum, an arming filament.′ nae. N.L. n. Devescovina, a genus of parabasalid flagellates; N.L. gen. n. devescovinae, of Devescovina (referring to the host genus).

Basis of assignment: filamentous rods, laterally attached on the surface of all Devescovina species, originally referred to as ‘Fusiformis-like rods’ by Kirby (1941). So far uncultured, but form a monophyletic group based on 16S rRNA gene sequence analysis (GenBank accession numbers: FN377751FN377758, AB194938, AB194939).

Experimental procedures


Termites originating from a broad geographic range were used (Table 1). Neotermes castaneus, Cryptotermes longicollis, Cryptotermes brevis, Cryptotermes dudleyi and Cryptotermes havilandi were obtained from cultures maintained at Federal Institute for Materials Research and Testing (BAM), Berlin, Germany. Cryptotermes secundus was collected in a mangrove forest near Darwin, Australia (provided by Judith Korb, University of Osnabrück, Germany). Cryptotermes cavifrons and Neotermes jouteli were collected in Florida, USA (provided by Rudolf H. Scheffrahn, University of Florida). If reference sequences were available, morphological identification of termites was confirmed by sequencing their COII genes using primers C2-J-3096 and TK-N-3807 described by Thompson and colleagues (2000); the COII genes sequences of all termites included in this study were submitted to GenBank.

Termites were maintained on dry wood at 27°C and 72% humidity. Colonies of N. castaneus and N. jouteli were occasionally given a few drops of water. To minimize the autofluorescence arising from cellulose in wood, termites used for hybridization experiments were fed crystalline cellulose powder for 3–4 weeks; in this case, all species required water. Only pseudergates were used in all experiments.

Scanning electron microscopy

The hindguts of four individuals of N. castaneus (BAM 2) were pulled out with fine forceps and opened in the fixative (2.5% glutaraldehyde in 0.1 M sodium phosphate buffer, pH 7.2). After fixation (30 min), samples were washed three times in buffer (15 min), post-fixed in 1% OsO4 in buffer (1 h, on ice), and washed again three times (15 min). After dehydration in a graded series of ethanol, cells were dried with a Balzer CPD 030, and coated with gold in a Balzer SCD 040. Flagellates were examined using a FEI Quanta 200 scanning electron microscope.

Flagellate picking and DNA extraction from flagellates

For SSU rRNA gene clone libraries of flagellate suspensions, the gut contents of one to three termites were suspended in Solution U (Trager, 1934). Using an inverted microscope, 50–200 cells of the respective morphotypes were collected by micropipette on Teflon-coated slides in a drop of phosphate-buffered saline (PBS; 0.13 M NaCl, 10 mM Na-phosphate; pH 7.4) and transferred two to three times into fresh PBS to remove loosely attached bacteria. The flagellate suspension was transferred to a polypropylene vial, boiled for 10 min at 95°C, and freeze-thawed on dry ice. DNA was extracted with the Nucleospin kit (Macherey-Nagel). Extracted DNA was directly used for amplification of SSU rRNA genes from flagellates and symbiotic bacteria. For the SSU rRNA gene clone libraries of whole termite gut homogenates, DNA was extracted as described previously (Ikeda-Ohtsubo et al., 2007).

PCR amplification and cloning

SSU rRNA genes of flagellates were amplified using the eukaryotic primers EUK19f (5′-AYYTGGTTGATYCTGCCA-3′) and EUK1772r (5′-CBGCAGGTTCACCTAC-3′) designed by Ohkuma and colleagues (1998). PCR products were purified using the MinElute purification kit (Qiagen) and were eluted in 10 μl elution buffer. Bacterial 16S rRNA genes were obtained with bacterial universal primers 27f (Edwards et al., 1989) and 1492r (Weisburg et al., 1991). PCR products were ligated into a plasmid pCR2.1-TOPO and introduced into Escherichia coli TOP10F′ by transformation using the TOPO TA cloning kit (Invitrogen). Transformants were checked by direct PCR using M13 primers, and correct-sized PCR products were subjected to ARDRA (enzymes MspI and HhaI). Several representatives of each ARDRA pattern were sequenced using M13 primer sets. Sequences newly obtained in this study have been submitted to GenBank under accession numbers FN377751FN377811.

Phylogenetic analysis

The SSU rRNA gene sequences were imported into the Silva database (Pruesse et al., 2007) implemented in the ARB software package (version 08; Ludwig et al., 2004). Automatic alignment of sequences was followed by manual refinement. The phylogenetic trees were calculated with the almost full-length SSU rRNA gene sequences (> 1300 bases) using ML (fastDNAml), maximum-parsimony (DNAPARS), and neighbour-joining (Jukes-Cantor correction) methods implemented in ARB. A 50%-consensus filter was used to exclude highly variable positions.

For phylogenetic analysis of termite COII genes, DNA sequences were imported into the ARB software package. Sequences were translated into amino acids using the invertebrate mitochondrial code, and translated amino acids were aligned. A 30%-consensus filter was used to exclude variable positions. Phylogenetic trees were calculated using a ML method (Protein_ML, amino acid substitution model Dayhoff), and bootstrap analysis was performed with phylip PROML, both implemented in ARB.

For cospeciation analysis of devescovinids and Bacteroidales ectosymbionts, tree topologies and node support were tested by additional analyses (ML, paup and Bayesian). The filtered alignment (ARB) was imported into paup (version 4.0b10; Swofford, 2002) and MrBayes (version 3.1.2; Ronquist and Huelsenbeck, 2003). Gaps in the alignment were treated as missing data. The model of nucleotide substitution for ML analysis selected by MODELTEST (version 3.7; Posada and Crandall, 1998) was GTR+I+Γ for both devescovinid and Bacteroidales alignments. Maximum-likelihood trees were inferred from heuristic searches under the Akaike information criterion, and a starting tree was generated by stepwise addition with 10 random replicates with TBR branch swapping. Nodal support was assessed by bootstrap analysis consisting of 100 bootstrap replicates, using the same heuristic search options. For Bayesian analyses, the substitution model for each alignment selected by MrModeltest (version 2.2; Nylander, 2004) was GTR+I+Γ for both flagellate and bacterial SSU rRNA gene alignments. For the 50% majority rule consensus trees, four Markov chains were simultaneously run for 1 000 000 generations, and parameters and trees were sampled every 100 generations. The consensus tree calculated from the 10 001 trees sampled after the initial burn-in period provided estimation of posterior probabilities.

The significance of the cospeciation events observed in the host and symbiont trees was assessed using the randomization test (1000 replicates) implemented in TreeMap (version 1.0a; Page, 1994).

Whole-cell in situ hybridization

Hindgut contents of N. castaneus were fixed with 3% (w/v) paraformaldehyde in PBS (0.13 M NaCl, 10 mM Na-phosphate; pH 7.4) for 2 h at 4°C. The cells were washed three times with ice-cold PBS, resuspended in PBS with an equal volume of ethanol, and stored at −20°C. The oligonucleotide probe DVB178 (5′-GCGGCTCCCCTGTTCTATC-3′), specific for the Bacteroidales ectosymbionts of Devescovina spp. in N. castaneus, was designed using the probe design functions of the ARB software (Ludwig et al., 2004) and checked using the Probe Match function of Ribosomal Database Project II ( It had two strong mismatches with all sequences in public databases. Optimal hybridization stringency was determined by varying formamide concentrations in the hybridization buffer over a range of 0–40% in 5% intervals at a fixed temperature of 46°C. Probe DVB178 hybridized exclusively with the target cells at any formamide concentration up to 20% (routinely used); at 25% formamide, the hybridization signal was only weak. Double hybridization with the Cy3-labelled probe DVB178 and the fluorescein-labelled general Bacteroidales probe CF319a was performed as described previously (Stingl and Brune, 2003), except that the ethanol series was omitted to minimize the distortion of flagellate cells. After hybridization and washing, the slides were quickly dried with compressed air and stained with 1 μg ml−1 4′,6-diamidino-2-phenylindole, washed with ice-cold 80% ethanol, air-dried, and covered with Citifluor (Citifluor). Samples were inspected by epifluorescence microscopy using an Axiophot microscope (Zeiss). Unspecific probe binding was checked by simultaneous hybridization with a fluorescein-labelled EUB338 probe (Amann et al., 1990) and a Cy3-labelled NON338 probe (Wallner et al., 1993).


Mahesh S. Desai received a doctoral fellowship from the International Max Planck Research School for Environmental, Cellular, and Molecular Microbiology. The authors are grateful to Judith Korb, Rüdiger Plarre and Rudolf H. Scheffrahn for providing termites and to Hans G. Trüper for etymological advice. We also thank Karen A. Brune for critical comments on the manuscript.