**3. Results**

## *3.1. Host Identification*

Identification of the host species was based on morphological and molecular features that included cell size, number of micronuclei and pores, structure of contractile vacuoles and the sequence of 18S rRNA gene. For morphological identification, we employed the key to the main morphospecies made by Fokin [36]. *Paramecium* cells of BMS16-23, BMS16-34 and BMS17-1 strains were preliminarily identified as *P. nephridiatum*, which is characterized as a 150–200 μm long cell, containing 2–6 micronuclei and from 2 to 5 pores of contractile vacuoles [36]. Subsequently, morphological identification was confirmed by direct sequencing of 18S rRNA gene amplified fragments of BMS16-23 and BMS16-34 cells. The obtained sequences (1748 bp and 1640 bp, accession numbers MK764889 and MK764894, respectively) showed 99–100% identity (NCBI GenBank) to the sequences of *P. nephridiatum* (MG573198.1, [49]; HE978251.1, [50]) and one sequence of *P. duboscqui* (HM140398.1; reference unpublished). Given clear morphological differences between these two species [36] and morphological similarity of the ciliates of our strains and *P. nephridiatum*, the obtained sequences were annotated as corresponding to the species *P. nephridiatum*. Interestingly, the comparison of the HM140398.1 sequence to other 18S rRNA of *P. duboscqui* showed only ~93% similarity implying that, apparently, the sequence HM140398.1 must have been derived from *Paramecium* species misidentified as *P. duboscqui*.

#### *3.2. Morphology and Biology of the Endosymbionts*

Two morphologically unlike bacterial endosymbionts were found in the BMS population of *Paramecium* (Figure 1). Two morphotypes could be easily distinguished on the basis of their morphology. The endosymbionts of the first morphotype were curve-shaped (CS), 0.6–0.8 μm wide and 4–7 μm long (Figure 1A,B). In turn, the bacteria of the second morphotype were rod-shaped (RS), 0.4–0.5 μm wide and 1.1–2.5 μm long (Figure 1C). Both CS- and RS-endosymbionts formed a triple symbiotic association (CS + RS + host) in cells of BMS16-23 and BMS17-1 strains, while the strain BMS16-34 harbored only RS-endosymbionts, thus representing a double symbiotic system.

**Figure 1.** Curve-shaped (CS)-bacteria from the cytoplasm of *P. nephridiatum* strains BMS16-23, BMS17-1, DIC; (**A**) a fragment of the living cell cytoplasm; white arrows point to CS-bacteria in the cytoplasm; (**B**) bacteria released from a squashed cell (arrows); (**C**) a fragment of the intact cell cytoplasm; the outlined area encloses a bacterial cluster consisting of CS-bacteria, white arrowheads point to rod-shaped (RS)-bacteria; double arrowheads point to structures resembling mitochondria; (**D**) the surface of the intact cell; white arrow points to the bacteria attached to the paramecia. Scale bar 5μm (**A**), 10 μm (**B**–**D**).

The CS-bacteria often formed clearly visible clusters in living host cells (Figure 1C). Apart from CS-endosymbionts, these clusters comprised host cytoplasm structures resembling mitochondria. In rare cases of hyperinfected cells, CS-bacteria were found sticking out of the cortex of the host cell (Figure 1D). However, it remains elusive whether this is a sophisticated mechanism of egress from the host cell or a consequence of cell overpopulation resulting in disintegration of the host pellicle. Microscopy of living and squashed host cells showed no motility of CS-bacteria, either in cytoplasm of a host cell or in cultivation medium. In contrast, RS-bacteria demonstrated fast chaotic movements both inside and outside of host cells resembling continuous tumbling motility.

The prevalence for both endosymbionts was 100% at the start of the laboratory culture; however, the number of CS-symbionts per host cell in BMS16-23 gradually decreased over time, and eventually, these endosymbionts were lost. Further work was continued using BMS17-1 strain. Nevertheless, by the end of the second year of cultivation, the prevalence for CS-symbionts in BMS17-1 also reduced to about 10% or less, and these symbionts seemed to have been lost in this strain as well. However, checking the culture over time revealed several ciliates still infected with CS-endosymbiont. Thus, the infection persists, though at a very low prevalence. During 2 years of observations, we regularly observed the fluctuations of both the prevalence and of the endosymbiont load per cell. On the contrary, the prevalence of RS-bacteria did not change with time in any of the three strains. Both endosymbionts resided exclusively in the cytoplasm and were never observed in nuclei or other compartments throughout the whole period of cultivation.

Killer-trait assay of BMS17-1 endosymbionts did not show any decrease in the naive cell number that would indicate a killer effect. However, no mating was done to exclude mate-killing ability. Inourexperiments,neitheroftheendosymbiontsdemonstratedanyinfectiouscapacity.

#### *3.3. Fine Structure of the Endosymbionts*

In TEM sections, the CS-symbionts showed the typical morphology of Gram-negative bacteria. The bacteria were not surrounded by an additional host membrane and lay "naked" in the cytoplasm (Figure 2A). The cytoplasm of CS-symbionts had moderate electron density with conspicuous ribosomes and free electron-lucid areas (Figure 2A).

**Figure 2.** Transmission electron microscopy images of CS-bacteria in the cytoplasm of *P. nephridiatum* strain BMS16-23. (**A**) The bacterium in the cytoplasm. (**B**) A fragment of a putative conglomeration with the host mitochondria. Scale bar 500 nm.

No other morphological structures were registered (Figure 2). In order to confirm our observations made with TEM, we examined the surface of bacteria with AFM (Figure 3). The images obtained with AFM did not show any structures on the bacterial surface, like flagella or an invasion tip, a peculiar feature typical for the infectious forms of *Holospora* [51] and other representatives of *Holosporales*, "*Ca*. Gortzia" and *Preeria* [9,52,53], which is thought to be crucial for the infection process of naive host cells in these species. It is noteworthy that the bacteria were often located close to mitochondria; however, direct contact of membranes was never observed (Figure 2B, Figure S1). This finding is consistent with live-cell observations and suggests that CS-bacteria tend to form clusters with mitochondria.

In contrast, the RS-bacteria were randomly distributed in the cytoplasm and never formed conglomerates with CS-symbionts or host cell organelles. In electron micrographs, RS-endosymbionts demonstrated two typical membranes of Gram-negative bacteria (Figure 4A). The cytoplasm was electron-dense and homogeneous without any inclusions. Bacterial cells were not enclosed by any additional membrane and were always surrounded by electron-lucid host cytoplasm lacking ribosomes. No flagella were observed in fine sections, however, negatively stained RS-bacteria showed moderate flagellation (up to seven flagella) (Figure 4B), which was confirmed by probing bacterial surface with AFM (Figure S2). The flagella were evenly distributed all over the bacterial surface, implying a peritrichous type of flagellation.

**Figure 3.** 3D reconstruction of CS-bacterium released from the cytoplasm of *P. nephridiatum* strain BMS16-23, AFM.

**Figure 4.** Transmission electron microscopy images of RS-bacteria in the cytoplasm of *P. nephridiatum* strain BMS16-34; (**A**) the bacterium in the cytoplasm; (**B**) the moderately flagellated bacterium cell, negative staining; white arrow points to the flagellum; (**C**) an enlarged fragment of bacterium surface; white arrowheads point to the hooks and l-rings. Scale bar 500 nm.

#### *3.4. Molecular Characterization of the Endosymbionts*

Nearly full-length sequences of 16S rRNA gene of endosymbionts of BMS16-23 (CS-symbiont: 1479 bp, accession number MK673804; RS-symbiont: 1478 bp, MK775140) and BMS16-34 (RS-symbiont: 1479 bp, MK775139) strains were obtained and compared with the data available from NCBI GenBank. The sequences of RS-bacteria from two strains were 99% similar and showed the highest identity score (99–100%) with the sequences of "*Ca.* Megaira venefica" (MG563925.1, MG563928.1) that has been described recently [49]. The highest sequence identity (95%) of CS-symbionts was observed with another *Paramecium* endosymbiont "*Ca.* Paraholospora nucleivisitans" (EU652696.1) [54]. Based on the obtained sequences, species-specific, fluorescently labeled oligonucleotide probes (16S\_Myst965 for CS-symbiots and 16S\_MegVen1226 for RS-bacteria (Table S1)) were designed to verify that the obtained sequences belonged to the endosymbionts (CS- and RS-bacteria, correspondingly) residing in the cells of BMS16-23, BMS16-34 and BMS17-1 strains. The designed probes were employed together with the Eub338 probe and were tested in FISH experiments at 0%, 15% and 30% formamide

concentrations. The 16S\_Myst965 probe showed specific hybridization with CS-symbionts at all formamide concentrations. Binding of the 16S\_MegVen1226 probe to 16S rRNA of RS-symbionts was observed at 0% and 15%, but not at 30% concentration of formamide. Surprisingly, the 16S\_MegVen1226 probe revealed non-specific labeling of CS-symbionts at 0% and 15% formamide. Increasing formamide concentration up to 20% led to specific labeling of the RS-symbionts only. Mapping the 16S\_MegVen1226 nucleotide probe to 16S rRNA gene sequence of BMS16-23 cells revealed that they share 17 of 22 nucleotides of the probe, which accounts for non-specific binding.

The FISH results demonstrated exclusively cytoplasmic localization of both endosymbionts (Figure 5). In accordance with live-cell observations and TEM images, CS-bacteria showed non-random distribution in the cytoplasm forming clusters and occasional location underneath the host cell membrane (Figure 5A–C). The bacterial load in a cell varied substantially from single cells to multiple cell clusters located throughout the whole cytoplasm (Figure 5). Moreover, bacteria penetrating the cortex were observed in some heavily infected ciliates (Video S1). On the contrary, RS-bacteria seemed to be uniformly distributed in the cytoplasm (Figure 5C,D).

**Figure 5.** Fluorescent in situ hybridization of the endosymbionts in cells of *P. nephridiatum*, Confocal Laser Scanning Microscope (CLSM); ( **A**,**B**) BMS16-23 and BMS17-1 cells labeled with 16S\_Myst965 probe (red) and the universal eubacterial probe Eub338 (green) counterstained with DAPI (cyan), FA concentration 15% and 30%, respectively; ( **C**) BMS17-1 cells labeled with 16S\_MegVen1226 (red) and 16S\_Myst965 (green) probes counterstained with DAPI (cyan), FA concentration 20%; ( **D**) BMS16-34 cells labeled with 16S\_MegVen1226 (red) and the universal eubacterial probe Eub338 (green) counterstained with DAPI (cyan), FA concentration 0%. Scale bar 10 μm.

Phylogenetic reconstructions covering 63 sequences of 16S rRNA gene including two sequences corresponding to identified RS- and CS-bacteria were made (Figure 6). The sequence of RS-symbiont formed a well-supported monophyletic clade with sequences of "*Ca.* Megaira venefica" and "*Ca.* Megaira polyxenophila" and formed one defined clade with sequences of pathogens *Rickettsia* and symbionts "*Ca.* Trichorickettsia" and "*Ca.* Gigarickettsia", order Rickettsiales. In turn, the sequence of CS-symbiont grouped with representatives of order Holosporales. In particular, it formed a branch distant to the core *Holospora* spp. and standing apart, like the sequences of other orphan taxa *Preeria caryophila* [53], "*Ca.* Paraholospora nucleivisitans" [54], and "*Ca.* Bealeia paramacronuclearis" [55]. Within this clade, CS-symbiont showed closer relation to another *Paramecium* endosymbiont "*Ca.* Paraholospora nucleivisitans". Of note, the *Holospora*-like bacteria tending to associate with the host energy organelles taken into analysis ("*Ca*. Cytomitobacter" spp., "*Ca*. Hydrogenosomobacter endosymbioticus" and the CS-symbiont, which we propose to name "*Ca*. Mystax nordicus") form separate clades in the phylogenetic tree, suggesting that the a ffinity to the host energy organelles may have originated independently.

**Figure 6.** Maximum likelihood phylogenetic tree of 63 sequences inferred on 1399 16S rRNA gene nucleotide positions with the GTR+I+G substitution model. Reported in bold are sequences characterized in this study. Numbers associated with each node correspond to maximum likelihood bootstrap values (values below 70% are omitted); numbers in brackets indicate the number of sequences representing that clade. The bar stands for an estimated sequence divergence of 4%. Abbreviations: *Ca.*—Candidatus; unc.—uncultured.
