Promiscuous and specific bacterial symbiont acquisition in the amoeboid genus Nuclearia (Opisthokonta)

(Opisthokonta) cyanobacteria ABSTRACT We isolated 17 strains of the amoeboid genus Nuclearia (Opisthokonta) from five Swiss lakes. Eight of these nucleariid isolates were associated with bacterial endosymbionts and/or ectosymbionts. Amoebae were characterized morphologically and by their 18S rRNA genes. Phylogeny based on molecular data resulted in four established monophyletic branches and two new clusters. A heterogeneous picture emerged by highlighting nucleariids with associated bacteria. Apart from one cluster which consisted of only isolates with and three groups of amoebae without symbionts, we also found mixed clusters. The picture got even more ‘blurred’ by regarding the phylogeny of symbiotic bacteria. Although seven different bacterial strains could be identified, it seems that we still are only scratching the surface of symbionts’ diversity. Furthermore, types of symbioses might be different depending on host species. Strains of Nuclearia thermophila harboured the same endosymbiont even when isolated from different lakes. This pointed to a specific and obligate interaction. However, two isolates of N. delicatula were associated with different endosymbiotic bacteria. Here the symbiont acquisition seemed to be rather promiscuous. This behaviour regarding symbiotic associations is especially remarkable considering the phylogenetic position of these basal opisthokonts.


INTRODUCTION
Intimate associations between unicellular or invertebrate eukaryotes and prokaryotes are ubiquitous, and their importance for the evolution of 'higher' life forms is increasingly recognized (Smith 1989;McFall-Ngai et al. 2013;Alegado and King 2014;Kiers and West 2015). We can intuitively argue that the probability of interactions increases if the spatial distance between hosts and potential symbionts is small, which is often the case for protists and bacteria. Knowing that such interac-tions are manifold, we use the term symbiosis in a very general way. We call the phenomenon of a close association of dissimilar organisms a 'symbiosis', thus we follow the original definition of this term by de Bary (see Appendix 1 in Paracer and Ahmadjian 2000). On an evolutionary scale, symbioses between eukaryotes and prokaryotes may emerge and disintegrate constantly and only a minute part will turn into 'stable associations'. The most stated and intensively studied examples are mitochondria and plastids that originated from the endosymbiosis of a host cell with alphaproteobacteria (Thrash et al. 2011) and cyanobacteria (Rodríguez-Ezpeleta et al. 2005), respectively. Beside the fundamental functions of respiration and photosynthesis, we know several traits which bacterial symbionts may provide to their eukaryotic hosts, e.g. they can be important for the host's nutrition, defence, competition and adaption to the environment (Gast, Sanders and Caron 2009). Associated bacteria can also be involved in the production (Freeman et al. 2012) or degradation of secondary metabolites including toxins (Kikuchi et al. 2012;Dirren et al. 2014).
Here we focus on members of the amoeboid genus Nuclearia (Opisthokonta, Nucleariidae) which can live in symbiosis with ecto-and endosymbiotic bacteria. Nuclearia is a single genus in the family Nucleariidae which is a sister group to Fungi (Zettler et al. 2001;Steenkamp, Wright and Baldauf 2006;Liu et al. 2009). As far as we know, there is only one documented case of an opisthokont protist with prokaryotic symbionts. Wylezich et al. (2012) described the choanoflagellate Codosiga balthica, which harboured two different endosymbiotic bacteria inside the cytoplasm. This lack of evidence is remarkable considering the importance of symbiotic interactions for multicellular opisthokonts. Nucleariid amoebae are usually surrounded by a glycocalyx (Moran, Gupta and Joshi 2011;Ouwerkerk, de Vos and Belzer 2013), which can be colonized by ectosymbiotic bacteria (Artari 1889;Cann and Page 1979;Patterson 1984;Cann 1986). In a previous study, we characterized Nuclearia sp. strain N (hereafter named Nuclearia thermophila strain N) which harboured the bacterial ectosymbiont (Paucibacter toxinivorans) nicely arranged inside the glycocalyx (Dirren et al. 2014). The interaction between N. thermophila strain N and this prokaryote seemed to be specific and stable.
Multicellular organisms usually are associated with more than one bacterial species. Ectosymbionts form entire assemblages which are designated as microbiota of the respective host. The microbiota of very 'simple' animals like the cnidarians Hydra (Fraune and Bosch 2007;Franzenburg et al. 2013) and corals (Lema, Bourne and Willis 2014) seem to be relatively distinct and even species specific. In higher animals including humans (Huttenhower et al. 2012), the microbiota is more diverse and variations between individuals within the same population are pronounced. However, in contrast to the taxonomic variability, the functional roles of such assemblages seem to be conserved. Thus, composition and function of the microbiota is essential for the organism's well-being. A multitude of diseases are consequently caused by regime shifts to unhealthy and unstable states (Lozupone et al. 2012). The 'simplicity' of the Nuclearia system could be a great benefit for the fundamental understanding of interactions of prokaryotes with their opisthokont hosts.
Symbiotic interactions of Nucleariidae are not restricted to ectosymbiotic associations but amoebae may additionally harbour bacterial endosymbionts. For example, N. radians (described as Nucleosphaerium tuckeri by Cann and Page 1979) may be associated with ectosymbiotic and endosymbiotic bacteria. In recent studies, the rickettsial endosymbiont of N. pattersoni (Dykova et al. 2003) and Candidatus Endonucleariobacter rarus (Dirren et al. 2014) of N. thermophila strain N were characterized. Endosymbiotic bacteria are not at all as common in higher life forms as ectosymbionts. The barrier for bacteria to enter metazoans' cells is rather rigid and well protected (e.g. by the immune system). In vertebrates, mainly pathogens are able to enter cells causing infections and pathological states (Casadevall 2008). From an evolutionary point of view, this is of great interest as multicellular organisms seem to 'outsource' their bacterial associations to preserve their integrity. Consequently, the glycocalyx can be regarded as a kind of 'external organ' harbouring the symbiotic assemblage. This arrangement not only allows benefiting from the microbiota but also ensures a minimal physical distance and thus protection (Fraune et al. 2015).
To sum up, from a phylogenetic perspective, Nucleariidae might be good model organisms to verify hypotheses about symbioses in general. In order to study these interactions, it is the first step to elucidate the diversity of Nucleariidae and to characterize in parallel their symbionts. In this study, (i) we report on the morphology and taxonomic affiliation (18S rRNA genes) of 17 Nuclearia strains; (ii) all isolates were screened for bacterial symbionts both in the glycocalyx and inside amoebae; (iii) finally, we focused on symbionts of N. delicatula and N. thermophila strains, and analysed the ultrastructure and the intracellular localization of endosymbiotic bacteria via transmission electron microscopy (TEM). Additionally, we sequenced the bacterial 16S rRNA genes for phylogenetic analyses.

Strains and cultures
A total of 17 Nuclearia strains were isolated from benthic and pelagic water samples of five Swiss lakes (Table 1). Single cells were picked with a glass pipette and washed in sterile water to generate monoclonal xenic amoebal cultures. Finally, isolates were cultured in autoclaved mineral water (Cristalp) and the cyanobacterium Planktothrix rubescens was added as sole food source. Planktothrix rubescens BC 9307 was isolated from Lake Zurich (Walsby, Avery and Schanz 1998) and is kept as axenic stock culture. Nuclearia cultures were maintained at a 12 h light (irradiance: 5-15 μmol m −2 s −1 )/12 h dark cycle in Tissue Culture Flasks 25 cm 2 (TPP) at 18 • C. Cultures were fortnightly renewed by adding 1 ml of the axenic cyanobacterial stock culture to 10 ml of new culture medium inoculated with 200 μl of an older culture. For all analyses, we included the dataset about N. thermophila strain N and its bacterial ectosymbiont P. toxinivorans strain SD41 (HG792253), originating from our previous study (Dirren et al. 2014). Nuclearia delicatula strain G (CCAP 1552/6), N. moebiusi strain K (CCAP 1552/7), N. thermophila strain N (CCAP 1552/5) and N. pattersoni strain A2 (CCAP 1552/8) were deposited in the Culture Collection of Algae and Protozoa (CCAP).

Morphological analysis and cladistic tree
Morphological characters were observed by light microscopy on living specimens. Features like multinucleate/uninucleate, evident/not evident nucleolus and branching of filopodia were observed when cells adapted a flattened form under the compression of the cover slip. Cells were considered 'spherical' if floating individuals in the water column could be observed (even if they were not always 'perfect' spheres). Strains were classified as being able to adapt a 'flattened form' when cells have ever attached to and moved on surfaces. The formation of syncytia was defined as the fusion of two or more cells. In addition, we checked all culture flasks for the appearance of cysts. The glycocalyx was either seen with phase contrast as translucent halo surrounding cells or after staining with Alcian blue. Ectosymbionts were defined as bacteria inside the glycocalyx located close to the cell membrane (loosely attached bacterial cells on the outer border of the glycocalyx were not classified as symbionts). Endosymbionts were detected with epifluorescence microscopy after DAPI staining and by in situ hybridization (CARD-FISH). The body diameter of spherical cells was If not stated otherwise, strains were isolated from benthic samples. Strains are listed according to the phylogenetic tree shown in Fig. 1A. In case that bacterial symbionts could be identified with CARD-FISH, the adequate oligonucleotide probes are listed (see Table 2  Amoebal strains were isolated from the pelagic zone of the respective lake. Downloaded from https://academic.oup.com/femsec/article-abstract/92/8/fiw105/2470017 by guest on 26 July 2018 measured more than 3 months after isolation of the strains. Only for N. thermophila strain D6, additional measurements were taken right after isolation. For calculations of the cladistic tree, morphological characters were judged as either present or absent and each strain was attributed to one of three size classes: 1. x < 13 μm; 2. 13 μm < x < 20 μm; 3. x > 20 μm. The cladistic tree (Jaccard's similarity coefficient) was calculated with the Add-In software XLSTAT (Addinsoft).

Sequencing of the 18S rRNA genes (Nucleariidae)
DNA was extracted from aliquots (1.5 ml) of Nuclearia cultures with the GenElute Bacterial Genomic DNA Kit (Sigma). PCR with GoTaq R Green Master Mix (Promega) and the eukaryote-specific primers Euk328f and Euk329r (Moon-van der Staay, De Wachter and Vaulot 2001) were used to amplify the 18S rRNA genes. If gel elecrophoresis resulted in the detection of bands of expected size, PCR products were purified with QIAquick PCR Purification Kit (Qiagen) and Sanger sequenced with ABI BigDye chemistry on an ABI 3130x Genetic Analyzer (Applied Biosystems). In order to sequence the entire amplicons, the additional primers SR2f, SR2r, SR4f, SR6f, SR6r, SR8f, SR8r, SR10f and SR10r (Nakayama et al. 1998) were used. In seven cases (strains G, D, S4, D4, B6, B1 and B3), the direct sequencing was not successful. Here 18S rRNA genes were again amplified from the extracted DNA with Platinum PCR Super Mix High Fidelity (Invitrogen) and the primers Euk328f and Euk329r. Subsequently, PCR products were purified as mentioned above and cloned into Escherichia coli using a pGem-T Vector (Promega) according to the manual. Clones were screened for expected size inserts with the plasmid primers M13f and M13r. Positive clones were grown in liquid cultures, and plasmids were purified with GenElute Five-Minute Miniprep Kit (Sigma). Inserts of plasmids were sequenced in the same way as PCR products but plasmid primers were used instead of Euk328f and Euk329r.

Sequencing of the 16S rRNA genes (symbionts)
Two 16S rRNA gene clone libraries were constructed from N. delicatula strain D and strain G, respectively. About 130 Nuclearia cells were picked with a micropipette and washed in sterile water. After three freeze-thaw cycles, DNA was extracted with GenElute Bacterial Genomic DNA Kit (Sigma). Extracted DNA served then as template for amplification of 16S rRNA genes with Platinum PCR Super Mix High Fidelity (Invitrogen) and the primers GM3f and GM4r (Muyzer and Ramsing 1995). After purification of PCR products and ligation into the pGem-T Vector (Promega), they were cloned following the manufacturer's protocol. Positive E. coli clones were detected by screening with plasmid primers (size ∼1.6 kbp) and their plasmids purified as described above. Sequencing of inserts was done with plasmid primers and the additional primer GM1f (Muyzer and Ramsing 1995). Partial 16S rRNA genes of the endosymbiont Candidatus Endonucleariobacter rarus from N. thermophila strain A and strain D6 were sequenced directly. The sequence of the probe CoNuc67 (Table 2) was used to design a specific primer (P1 CoNuc f 5 -TAACAGAGTGTGTAGC-3 ). PCR amplification with GoTaq Green Master Mix (Promega) was done with extracted DNA from these cultures using the forward primer P1 CoNuc f and the reverse primer GM4r (positive control: ext. DNA from N. thermophila strain N; negative control: ext. DNA from N. thermophila strain B1). Subsequently purified PCR products were directly sequenced with the primers P1 CoNuc f and GM4r as described above (LN875086-LN875088).

Phylogenetic analyses
The software DNA Baser v3.5.0 (Heracle BioSoft) served as tool for assembling partial sequences. Chimeric sequences were detected and removed using Mallard and Pintail (Ashelford et al. 2005). For phylogenetic analyses, the ARB software package (Ludwig et al. 2004) with the SILVA database SSU Ref 111 (Pruesse et al. 2007) was used. All available Nuclearia 18S rRNA gene sequences from described species, our isolates, and as outgroup two sequences from Candida sp. (AB013586 and EU348785) were included for phylogenetic tree reconstruction. Sequences were trimmed and aligned with the MAFFT aligner (Katoh and Standley 2013). Alignments were manually optimized and ambiguous regions (e.g. insertions in the V4, V7 and V8 domains) were removed resulting in 1501 positions with 150 distinct alignment patterns. Another phylogenetic tree including all our N. delicatula clones, N. delicatula (AF349563) and N. simplex (AF349566)/N. moebiusi (AF349565) as outgroup was calculated. In addition, sequences were aligned and trimmed as described above but none of the hypervariable regions were removed (2363 positions with 204 distinct alignment patterns).
For the reconstruction of phylogenetic trees, maximum likelihood (ML) and Bayesian inference (BI) methods were used. Bootstrapped ML trees were calculated (1000 iterations) using the RAxML algorithm (Stamatakis, Hoover and Rougemont 2008). The parameters were GTR (general time reversible) model with a distribution for rate heterogeneity among sites, with four discrete rate categories. BI was calculated using the Ex-aBayes software package ( C The Exelixis Lab). The posterior probabilities from BI trees (four chains; 100 000 generations) were added to ML trees where trees of both methods were congruent. Full-length 16S rRNA gene sequences from clone libraries and 18S rRNA gene sequences of the Nuclearia strains were deposited in the EMBL database with the accession numbers LN875040-LN875170.

CARD-FISH and probe design
First, CARD-FISH with the general probes EUB I-III (Daims et al. 1999), ALF968 (Neef 1997), BET42a, GAM42a (Manz et al. 1992), CF319a (Manz et al. 1996), HGC69a (Roller et al. 1994) and VER47 (Buckley and Schmidt 2001) allowed for the identification of symbionts on a higher taxonomical level. Afterwards clusters of potential symbionts were chosen from phylogenetic trees of the 16S rRNA gene clone libraries. Specific probes were designed based on the sequences of these candidate clusters. Probe design with the dedicated ARB tool resulted in four specific CARD-FISH probes ( Table 2). The Ribosomal Database Project (www.rdp.cme.msu.edu) and the web tool Mathfish (Yilmaz, Parnerkar and Noguera 2011) were used for in silico testing of the new probes. Appropriate formamide concentrations (for highest stringency) were determined empirically. Non-specific staining was addressed with the probe NON338 (Wallner, Amann and Beisker 1993). CARD-FISH on filters was done with differently la-belled tyramids (fluorescein and Alexa546) following the previously published protocol (Dirren et al. 2014). In addition, CARD-FISH of amoebae on gelatine-coated glass slides and embedded in agarose were prepared.

Microscopy and photographic documentation
Differential interference and phase contrast images were acquired with a Canon EOS1000D controlled by the software Ax-ioVision 4.8.2 (Zeiss) installed on an Axio Imager.M1 microscope (Zeiss). CARD-FISH preparations were analysed at the same microscope with epifluorescence microscopy (Zeiss optical filter sets: set 01, 10, 14 and 43) and by confocal laser-scanning microscopy (SP5-R, Leica Microsystems, Germany).

Transmission electron microscopy
Glutaraldehyde (final conc. 1.25 %) and osmium tetroxide (final conc. 1 %) were mixed and added to small volumes of N. delicatula strain D and strain G cultures (after centrifugation at 1000 g for 20 min and discarding of supernatants). Fixation was done on ice for 1 h followed by two washing steps (centrifugation for 10 min at 2000 g and exchanging of fixative solution with H 2 O). Washed pellets were resuspended in melted agar (2 %). After hardening and cutting the agar block into smaller pieces (∼10 mm 3 ), they were block stained with uranylacetat (1 %) for 1 h at room temperature. Subsequently, samples were dehydrated in an ethanol series (70%, 80%, 96% and 100%) and finally in propylenoxide, followed by embedding in epon-araldite. Ultrathin sections were cut with an Ultracut UCT (Leica) and poststained with lead citrate for 6 min. An electron microscope Philips CM100 equipped with a digital camera (Gatan Orius 1000) was used for the analysis of the TEM preparations.

Morphological versus molecular phylogeny of Nuclearia spp.
For a long time, nucleariid amoebae were described using only morphological characters. However, the majority of these amoeboid species share many features used for their identification (see table 1 in Yoshida, Nakayama and Inouye 2009). It seems that some inadequately defined characters were even interpreted differently by researchers, e.g. if amoebae may form flattened/spherical cells, if a glycocalyx is present or absent and if the nucleolus is evident. This becomes obvious, when comparing fig. 11 in Patterson (1984), where the author stated the lack of a glycocalyx, with fig. 6 in Pernin (1976), where the presence of EPS was proven. In addition to these 'vague' characters, other features like the formation of multinucleate syncytia, cyst production and the appearance of branched filopodia might be rarely or not at all observed depending on culture and observation conditions. Even the cell size of single isolates may vary depending on culture conditions. We documented at least for one isolate a shrinkage of cells in the course of cultivation. The mean cell size of N. thermophila strain D6 decreased from initially 24.6 μm (day 7 after isolation, n = 56) to 15.7 μm (day 90, n = 100). Thus, most probably these inconsistencies and different interpretations of features led to redescriptions of species and incorrect identifications.
This assumption is additionally supported by the fact that two N. simplex isolates clustered in the 18S rRNA gene-based phylogenetic tree very distantly (Fig. 1A) with N. moebiusi and N. pattersoni, respectively. Because of this discrepancy, we considered the two as N. simplex identified isolates to belong to different species. Consequently, we will name these phylogenetic groups as N. moebiusi and N. pattersoni cluster, respectively. Moreover, two sequences of one and the same N. moebiusi isolate (AF349565 and AF484686) were included in phylogenetic trees by some authors (Dykova et al. 2003;Yoshida, Nakayama and Inouye 2009) which further caused confusions.
Nevertheless, in this study we partly worked with traditional morphological features for comparisons of our strains (Fig. 2 and Fig. S1, Supporting Information) with published species descriptions. To characterize our Nuclearia spp. isolates, we even included four additional features: benthic isolate, pelagic isolate, presence of ectosymbionts and presence of endosymbionts (Table 1). We checked if the morphological classification corresponded to the molecular phylogeny by performing a cladistic analysis based on presence/absence of characters. The cladistic tree (Fig. 1B) and the 18S rRNA gene-based ML tree (Fig. 1A) were in good accordance regarding the N. delicatula and the N. thermophila clusters. In both trees, they were sister groups including same isolates. In contrast, the third big cluster in the cladistical tree unified isolates from distant branches of the ML tree. Although the substructure of this third cluster reflected quite well the 18S rRNA gene-based phylogeny, two isolates clustered differently. In the cladistic tree, Nuclearia sp. strain NZ had no close relative and strain K formed together with strains A1 and D1 a group, which was not confirmed by molecular phylogeny. Taken together, only N. delicatula and N. thermophila strains could be identified solely by their morphological traits. For the affiliation of all other isolates, additional molecular information (18S rRNA genes) was needed.

Assignment of isolates to described species
The species descriptions of N. delicatula from Patterson (1984) and Cann (1986) are in good accordance with morphological features (Table 1) observed for all isolates in the N. delicatula cluster.
The morphological features described for N. moebiusi differed from what we observed for Nuclearia sp. strain K. We found spherical cells as well as a glycocalyx which was not reported for N. moebiusi. When considering the 'excavate cavities' ( fig. 11 from Patterson 1984) to be the glycocalyx and additionally taking the trait 'spherical form' less restrictive, we can assign our isolate to this species (i.e. N. moebiusi strain K).
The morphological incongruences between N. thermophila strain N and the original species description of N. thermophila by Yoshida, Nakayama and Inouye (2009) were discussed in our previous study (Dirren et al. 2014). For the isolates clustering together with N. thermophila, we reported a good accordance with the characters earlier described for N. thermophila strain N. Only the formation of syncytia could not be documented for strains B1 and D6. The present results including morphological and phylogenetic analyses of four different isolates (strains B1, D6, A, N) thus justify their assignment to the species N. thermophila.
All described features for N. pattersoni (Dykova et al. 2003) except the presence of endosymbionts could be observed for Nuclearia sp. strain B4 and A2. Taking their phylogenetic close relatedness (Fig. 1A) into account, we can assign them to the species N. pattersoni.
The isolates Nuclearia sp. strain NZ, strain A1 and strain D1 built a sister group to the N. pattersoni cluster (Fig. 1A). Morphologically these three isolates were very similar and the lack of cysts was the only character differentiating them from N. pattersoni isolates. However, the phylogenetic distance (Fig. 1A) still does not allow for assigning them to the described species.
Finally, the two isolates Nuclearia sp. strain B3 and A5 formed a discrete new phylogenetic group. In spite of only minor morphological differences (e.g. formation of syncytia in strain A5) to isolates in the N. moebiusi and N. pattersoni clusters, we suppose that they form a new species. Phylogenetic reconstruction even points at a rather basal position, probably representing a sister group to N. delicatula, N. moebiusi and N. thermophila. We added hypothetical gains and losses of the features 'cyst production' and 'formation of syncytia' to the corresponding branches (Fig. 1A). The fact that we found cyst production for isolates all over the tree may indicate that the common ancestor was encysting. In contrast, the formation of syncytia was found only for isolates in one of the main branches and thus could be an acquired trait.

Variability of the nucleariid 18S rRNA gene copies
For all but seven Nuclearia isolates, PCR amplification of the 18S rRNA gene with general eukaryotic primers and direct sequencing was successful (LN875106-LN875114). In contrast, assembling of partial 18S rRNA gene sequences failed for N. thermophila strain B1, Nuclearia sp. strain B3 and all N. delicatula isolates (strains G, D, S4, D4 and B6). For N. thermophila strain B1, a poly-G region causing 'hard stops' during sequencing resulted in two non-overlapping partial sequences. The first part (∼700 nt) and the second part (∼1230 nt) of the 18S rRNA gene could thus not be assembled. Sequence qualities of all N. delicatula strains and Nuclearia sp. strain B3 dropped in regions containing homopolymers due to superposition of signals. This pointed to sequence variations in multiple 18S rRNA gene copies (e.g. different lengths of homopolymers). Therefore, PCR products of N. delicatula strains and Nuclearia sp. strain B3 were cloned and de novo sequenced resulting in partial sequences with highquality scores even for regions containing homopolymers. Two to ten different clones were completely sequenced (LN875123-LN875170) for N. delicatula isolates and Nuclearia sp. strain B3. In order to exclude that interclone variation was introduced by PCR and sequencing errors, we reamplified 18S rRNA genes from cleaned-up plasmids of three N. delicatula strain D4 clones. The obtained sequences were identical to those generated by direct sequencing of inserts. Thus, detected interclone variations most probably originated from natural variations in 18S rRNA gene copies and were not artefacts.
Microheterogeneities in the nucleariid 18S rRNA genes have been already documented by Zettler et al. (2001). They mainly originate from size variations in the insertions inside the V4, V7 and V8 domains (sensu De Rijk et al. 1992). Pairwise sequence distances were calculated for each clone library of N. delicatula strains (G, D, S4, D4), Nuclearia sp. strain B3 and N. thermophila strain B1 (Fig. S2A, Supporting Information). Variations in the 18S rRNA gene copies of N. thermophila strain B1 (mean ± standard deviation: 0.21±0.07 %) were lower than those in N. delicatula strains (0.48±0.15 % to 0.63±0.24 %) and in Nuclearia sp. strain B3 (0.44±0.24 %). Intrastrain variations (distances of clone sequences: 0.58±0.05 %) were about three times higher than interstrain variations (distances of the consensus sequences: 0.17±0.07 %) for N. delicatula isolates. Thus, they could not be separated phylogenetically on the base of this marker gene (Fig.  S2B, Supporting Information). When we calculated sequence similarity of N. delicatula (AF349563) and N. delicatula strain G without these variable parts in the V4, V7 and V8 domains, we got a high value of 99.7 %. In contrast, sequence similarity including the hypervariable stretches was only 94 %. In the same way, 18S rRNA gene copies in single isolates are mainly diverging (e.g. due to insertion and deletion of nucleotides) inside the homopolymers of hypervariable domains. Slipped-strand mispairing (Levinson and Gutman 1987) might be the mechanism behind this phenomenon. A slightly higher mutation rate could also be detected for the variable stretches in sequences from the N. pattersoni cluster (e.g. sequence similarity of the described N. pattersoni and N. pattersoni strain B4: with homopolymer region 99 % and without 99.2 %) but not within the N. thermophila cluster. The sequence similarity of N. thermophila (AB433328) and N. thermophila strain A was 99.6% with and without homopolymer regions.
Taken together, divergence and/or number of 18S rRNA gene copies vary between different Nuclearia species. Regarding their mutation rates, homopolymer regions can differ drastically from the rest of the sequence. And finally, accumulations of mutations in these regions seem to be species specific.

Associations of Nuclearia spp. with prokaryotes
In total 8 of our 17 isolates were associated with endosymbiotic and/or ectosymbiotic bacteria. Symbionts could be detected for all N. delicatula strains but not for Nuclearia isolates from three other clusters (Fig. 1A). Beside these homogeneous branches, also mixed groups were found. In the N. thermophila cluster, three out of five representatives had symbionts (Fig. 1A). The N. pattersoni cluster was also heterogeneous. It was only reported for N. pattersoni (Dykova et al. 2003) that this amoeba harboured a rickettsial endosymbiont.
The non-systematic appearance of symbiotic associations inside the genus Nuclearia indicates a species-dependent disposition. As far as we know such a high variability within a single genus has been described only for Acanthamoeba spp. (Fritsche et al. 1993;Horn et al. 1999;Horn 2008). Either some Nuclearia species evolved traits by which the probability to enter a symbiotic relationship increases or it is a plesiomorph character that has been partly lost. Considering their phylogenetic position within opisthokonts, it is of interest to search specifically for such traits in future genetic analysis. Probably, nucleariids have already specific genes and machineries which are involved in selecting and controlling of symbiotic partners (Bosch 2014). The question about the frequency of prokaryotic symbionts in unicellular opisthokonts still remains to be addressed. Today, it is not clear if the lack of knowledge simply derives from the low number of studies looking for symbiotic associations or if the highly diverse interactions inside the genus Nuclearia are an exceptional phenomenon.

Nuclearia thermophila isolates and their ectosymbionts
In a previous study (Dirren et al. 2014), we identified the ectosymbiont of N. thermophila strain N as the betaproteobacterium  Table 2. Different fluorophore-specific filter sets were used to image the cells after hybridization. The first two and the last two columns of each row represent always the same cell. P. toxinivorans (Rapala et al. 2005) and designed the specific CARD-FISH probe 'Pauci995' (Table 2). Within the N. thermophila cluster, only strain A of our new isolates was also associated with ectosymbiotic bacteria right after isolation ( Fig. 2A). Unfortunately during cultivation, these bacteria got lost (Fig. 2B) before CARD-FISH filters could be prepared. Although cells were surrounded by a glycocalyx, no ectosymbionts were observed for strain D6 ( Fig. 2C; Fig. S1C, Supporting Information) and strain B1 (Fig. 2D). Since the ectosymbiont of strain N (P. toxinivorans strain SD41) was available as pure culture, we checked if the other strains could be infected with these bacteria. When we added an aliquot (1 ml) of a pure bacterial culture to the medium of ectosymbiontfree isolates, we observed a colonization of strains A and D6 (Fig. 3A-D). Surprisingly, this was not the case for strain B1. This experiment indicated a highly specific interaction of P. toxinivorans strain SD41 with only one phylotype of the N. thermophila cluster (strain D6, A and N; see Fig. 1A). The fact that the ectosymbiont did not colonize the glycocalyx of the close relative strain B1 points to a distinct contribution of the host to this symbiosis.
The importance of a glycocalyx has been most extensively investigated for epithelial cells of the gastrointestinal tract (Moran, Gupta and Joshi 2011). The composition of glycoproteins produced by the host determines the physical (e.g. viscosity) and chemical (e.g. site for bacterial adhesion) nature of this extracellular structure and thus the interaction with bacteria. On the other hand, the composition of the glycocalyx can be modulated by bacteria in distinct ways (Hooper and Gordon 2001). This suggests a cross-talk between host and bacteria mediated by the glycocalyx. Furthermore, in the early branching metazoan Hydra, receptors, species-specific antimicrobial peptides (Bosch 2014) and even viruses (Bosch, Grasis and Lachnit 2015) have been shown to be main factors shaping the ectosymbiotic bacterial community. Unfortunately, the molecular interactions between the N. thermophila strains and P. toxinivorans are yet not studied.

Nuclearia thermophila isolates and their endosymbionts
In the N. thermophila cluster, three (A, D6 and N) out of four isolates harboured the gammaproteobacterial endosymbiont Ca. Endonucleariobacter rarus (Fig. 4A-H and Dirren et al. 2014). Hybridization with the specific probe CoNuc67 (Table 2) resulted in positive signals from all bacteria in strains A and D6 (Fig. 4C, G-H). In contrast, endosymbionts were missing in strain B1 ( Fig. 3E and F) which additionally had a slightly divergent 18S rRNA gene sequence (Fig. 1A).
Interestingly, 18S rRNA gene sequences of strains D6, A and N were identical, but 16S rRNA gene sequences of their endosymbiont Ca. Endonucleariobacter rarus were slightly different. Endosymbionts of strains D6 and A formed a sister group to bacteria of strain N (Fig. 5A), although strains A and N were isolated from the same lake, and strain D6 from a 25 km distant lake.

Nuclearia delicatula isolates and their ectosymbionts
Four of our N. delicatula strains (D, G, S4 and D4) had both ectosymbionts (Figs 2E-I and 4Q-T) and endosymbionts (Figs 3I-P and 4I-P). Strain B6 was only associated with ectosymbionts . Based on the 16S rRNA gene clone library of strain G (Fig. S3A, Supporting Information), three specific oligonucleotide probes were designed: Bu154, Le827 and Del1424 ( Table 2). Ectosymbionts of strain G could be hybridized with the betaproteobacterial probe Bu154 (Fig. 4Q-T). The closest described relative (98.6% sequence similarity) to the cluster covered by this probe was Inhella inkyongensis (Song et al. 2009). Phylogeny of nucleariid's ectosymbionts (Fig. 5B) highlights that these ectosymbiotic bacteria are related (95.5% sequence similarity) to the earlier identified ectosymbiont of N. thermophila strain N (P. toxinivorans). Inhella sp. and P. toxinivorans have both sequence divergences to the bacteriochlorophyll a containing bacteria Roseateles (Suyama et al. 1999) and Rubrivivax (Willems, Gillis and De Ley 1991) of ∼4% and ∼5%, respectively. They form a metabolically diverse group sometimes referred to as 'Sphaerotilus-Leptothrix group' (Spring 2006;Song et al. 2009) inside the family Comamonadaceae. As far as we know, a symbiotic live style has not been reported for any representative of this group.
Nucleariid amoebae are conspicuous concerning their nutrition: they can feed on harmful filamentous cyanobacteria, without being affected by toxic secondary metabolites (Dirren et al. 2014). In the previous study, we showed that P. toxinivorans was able to degrade microcystins, the cyanobacterial toxins stored in food organisms. In the case of Inhella sp., we have yet no proof for any similar metabolic capability. However, the spatial proximity to the host's cell surface suggests an exchange of metabolites between the symbiotic partners.

Nuclearia delicatula isolates and their endosymbionts
The gammaproteobacterial probe Le827 and the deltaproteobacterial probe Del1424 gave positive CARD-FISH signals for intracellular bacteria of N. delicatula strain G and no signals from bacteria in the cultivation medium. Endosymbionts hybridized with probe Le827 specific for a cluster of gammaproteobacteria were evenly distributed and represented a small part of total bacteria inside the cells (Figs 3I-J and 4P). Because of the homogenous distribution and their estimated abundance by CARD-FISH, we could assign this phylotype to distinct morphological features (morphotype 1) observed on TEM pictures ( Fig. 6A and B). Bacteria had two membranes of a typical Gram-negative cell wall and an electron dense spot inside cells ( Fig. 6C and D). They were localized in the cytoplasm and mostly surrounded by an electron translucent halo but never by an additional membrane. Some intracellular bacteria observed in N. radians display remarkable morphological similarities to bacteria in N. delicatula strain G (see Plate 4c from Cann and Page 1979). No described relatives of our endosymbiotic bacteria could be found in public databases. Apart from some sequences of uncultured gammaproteobacteria (highest sequence similarity 94.3%), the closest relatives were Candidatus Berkiella aquae (88.5% sequence similarity) and Candidatus Berkiella cookevillensis (88.1% sequence similarity) (Fig. 5C). These bacteria were found after infection inside the nucleus of Acanthamoeba polyphaga (Mehari et al. 2016). We never detected bacteria inside the nuclear membrane of strain G and endosymbionts differed morphologically from the recently characterized symbionts (Mehari et al. 2016, e.g. no electron dense spot). The 16S rRNA gene sequences of these symbiotic bacteria and strain G's endosymbiont are too much diverged to resolve their phylogenetic relationship based solely on this marker gene. Thus, corresponding branches had to be collapsed (low support values) in the phylogenetic tree (Fig. 5C). We propose the taxonomic status 'Candidatus Ovatusbacter abovo' for the endosymbiont of N. delicatula strain G.
Double hybridization with the gammaproteobacterial probe Le827 and the deltaproteobacterial probe Del1424 showed different endosymbionts in strain G being hybridized (Fig. 4P). In contrast to the even distribution and low frequency of bacteria labelled with Le827, endosymbionts hybridized with Del1424 showed a lumped occurrence and were highly abundant (Figs 3K-L and 4P). These characteristics corresponded to the other prominent morphological phenotype (morphotype 2) seen on TEM pictures ( Fig. 6A and B). Cells had again a typical Gram-negative cell wall structure ( Fig. 6E and F) but in contrast to 'Candidatus Ovatusbacter abovo' they were always surrounded by an additional host-derived membrane. Small vacuole-like structures harboured single cells (Fig. 6G and H) or multiple bacteria of morphotype 2 ( Fig. 6E and F). We even detected these endosymbionts inside food vacuoles, often attached to the membrane of the vacuole, together with remnants of the food organism P. rubescens (Fig. 6B, G and H). In contrast to cyanobacterial cells, endosymbiotic bacteria seemed to be resistant to  digestion. This observation in combination with the fact that the deltaproteobacterial probe Del1424 did not hybridize with bacteria in the cultivation medium speaks against a possible role of these intracellular bacteria as food.
Usually bacterial pathogens are taken up by phagocytosis and then either prevent the fusion of lysosomes (e.g. Legionella pneumophila; Roy and Kagan 2000) or escape the phagosomes (e.g. Rickettsia prowazekii; Whitworth et al. 2005). Because of the facts that cyanobacterial cells were digested in food vacuoles and endosymbionts were never seen freely in the cytoplasm, there is no indication for any of these two strategies. TEM pictures of A. castellanii infected with the pathogenic symbiont The less abundant morphotype 1 can be found freely inside the cytoplasm, whereas the prominent morphotype 2 is present in vacuole-like structures (v). (C) Two bacteria and (D) a dividing individual of morphotype 1 inside the nucleariid cell. A characteristic central electron-dense spot and a typical Gram-negative cell wall structure with two membranes (arrowheads) are visible. No peribacterial membrane is present, but an electron translucent halo surrounds the cell. (E and F) Cells having the characteristics of morphotype 2 are tightly packed inside a peribacterial membrane (arrow). The Gram-negative cell wall organization with two membranes (arrowheads) is visible. The cell content of this endosymbiont has a homogenous appearance on TEM pictures. (G) In the big central food vacuole (see overview B), an intact P. rubescens filament (p) and remnants of digested cyanobacteria can be seen. In addition to the food organism, bacteria of the morphotype 2 are present inside the food vacuoles (v). Many bacteria seem to be attached to the membrane. Additionally, single or few cells of this morphotype are enclosed in membranes apparently not connected to the food vacuole. (H) Higher magnification of the interface between food vacuole and cytoplasm. Bacteria of the morphotype 2 are attached to the membrane sometimes forming cavities. Bacteria seem to be intact and not digested. d, dictyosome; f, filopodium; g, glycocalyx; m, mitochondrion; no, nucleolus; p, P. rubescens filament; v, vacuole-like structures. Scale bars represent 10 μm in (A and B), 4 μm in (G), 1 μm in (H) and 500 nm in (C-F).
'Candidatus Jidaibacter acanthamoeba' ( fig. 1 in Schulz et al. 2015) resemble conspicuously our observations of the frequent endosymbiont. In contrast to this accordance, types of the symbioses seem to differ. The regular exponential growth of the host N. delicatula speaks against a severe pathogenic nature of its endosymbiotic bacteria. Again, no close relatives of this endosymbiont belonging to the deltaproteobacteria were found in public databases. The closest relative (89.8% sequence similarity) was a pathogenic bacterium of daphnids named Spirobacillus cienkowskii (Rodrigues et al. 2008). None of the sequences included in the phylogenetic tree (Fig. 5C) clustered together with this endosymbiont of strain G. Thus, we propose the taxonomic status 'Candidatus Turbabacter delicatus' for these bacteria. Specific hybridization with the deltaproteobacterial probe Del1424 showed that the endosymbiont was also present in cells of strains S4 and D4 ( Fig. 3M-P). Like in strain G, not all of the intracellular bacteria were labelled. In contrast to the positive hybridization with the gammaproteobacterial probe Le827 with the other part of endosymbionts in strain G, no signal was obtained for strains S4 and D4. Most probably, these strains