Genome expansion of an obligate parthenogenesis-associated Wolbachia poses an exception to the symbiont reduction model

Background Theory predicts that dependency within host-endosymbiont interactions results in endosymbiont genome size reduction. Unexpectedly, the largest Wolbachia genome was found in the obligate, parthenogenesis-associated wFol. In this study, we investigate possible processes underlying this genome expansion by comparing a re-annotated wFol genome to other Wolbachia genomes. In addition, we also search for candidate genes related to parthenogenesis induction (PI). Results Within wFol, we found five phage WO regions representing 25.4% of the complete genome, few pseudogenized genes, and an expansion of DNA-repair genes in comparison to other Wolbachia. These signs of genome conservation were mirrored in the wFol host, the springtail F. candida, which also had an expanded DNA-repair gene family and many horizontally transferred genes. Across all Wolbachia genomes, there was a strong correlation between gene numbers of Wolbachia strains and their hosts. In order to identify genes with a potential link to PI, we assembled the genome of an additional PI strain, wLcla. Comparisons between four PI Wolbachia, including wFol and wLcla, and fourteen non-PI Wolbachia yielded a small set of potential candidate genes for further investigation. Conclusions The strong similarities in genome content of wFol and its host, as well as the correlation between host and Wolbachia gene numbers suggest that there may be some form of convergent evolution between endosymbiont and host genomes. If such convergent evolution would be strong enough to overcome the evolutionary forces causing genome reduction, it would enable expanded genomes within long-term obligate endosymbionts. Electronic supplementary material The online version of this article (10.1186/s12864-019-5492-9) contains supplementary material, which is available to authorized users.


Background
Endosymbiotic bacteria are found in the cells of many eukaryotic species, where they affect major processes such as host reproduction [1,2], defence against pathogens [3] and development [4,5]. Endosymbionts are dependent on their host for proliferation, whereas host are not necessarily dependent on their endosymbiont and can often reproduce and survive in their absence.
However, there are ample examples of obligate endosymbionts, e.g. Buchnera in aphids [6,7], Wigglesworthia in the tsetseflies [8], and Wolbachia in parasitic filarial nematodes [9,10]. In these cases neither host nor endosymbiont are viable without the other, and these associations are usually characterised by a long evolutionary history, nutritional or developmental dependency, and vertical transmission of the symbiont [11].
Current theory predicts that mutual dependency between host and endosymbiont leads to a reduction in genome size of endosymbionts [12,13]. Different genetic mechanisms have been implied as the driving forces behind these reductions. First, co-adaptation between the symbiont and their host may result in redundancy of certain symbiont functions releasing genes from selective constraints, causing them to decay and eventually to disappear [14,15]. Second, small population sizes and limited opportunity for horizontal gene exchange increases genetic drift to a level that purifying selection cannot overcome, which leads to deleterious mutations that cause pseudogenization of mildly advantageous genes and the eventual removal of those genes due to the inherent deletion bias found for bacterial genomes [12,16,17]. In addition, multiple transitional events seem to pinpoint moments of sever genome reduction [18]. These include becoming host bound, moving into a specialized host cell and being vertically transmitted through host generations. Intervals between different transitional events seem prone to different sets of selection pressures that influence genome size. For example, the negative correlation between host dependence and symbiont genome size only holds for vertically transmitted endosymbionts [19]. However, there is still a large amount of variation in the genome size of vertically transmitted endosymbionts that are not residing in specialized host cells, which is difficult to explain in the context of current theory [18]. Examining endosymbiont genomes in this stage will be needed to better understand all factors that influence the genome reduction during the evolution of endosymbionts.
Wolbachia is one of the most widespread endosymbionts [20,21] and known for its variety of interactions with its host, including male killing, feminization, cytoplasmic incompatibility (CI), parthenogenesis-induction (PI) and provisional mutualisms [22,23]. Plus, all known Wolbachia strains are in a similar transitional stage according to the classification of Lo et al. (2016), i.e. they are mainly transmitted vertically and do not reside in specialized structures. They are thus predicted to vary in genome size depending on the host dependency [18]. This prediction seems to hold for most of the Wolbachia strains, with reduced genomes that lack mobile elements found in obligate mutualistic strains and larger genomes in facultative reproduction-manipulating strains. However, recently a Wolbachia genome was sequenced that poses an exception to this pattern. The Wolbachia from the parthenogenetic collembolan host Folsomia candida has the largest Wolbachia genome sequenced so far [24], yet it is obligate to its host. Since, elimination of Wolbachia by heat or antibiotics, renders the host's eggs non-viable [25,26]. In addition, the wFol genome is the first fully sequenced genome from supergroup E that is positioned to be a sister group to supergroup A, B, C, D, F and H, and shares a more ancestral common ancestor with supergroup L [27,28]. All complete genomes from members of these supergroups have smaller genomes than wFol, which suggests that this obligate Wolbachia strain showed genome expansion rather than the reduction that would be predicted by the symbiont genome reduction model. Furthermore, the wFol genome could contribute to understanding the genes and mechanisms underlying PI, as wFol has been suggested to cause parthenogenesis in its host. All parthenogenetic F. candida lines carry Wolbachia whereas the sexually reproducing lines do not [25,29,30]. Unequivocal proof of PI would require curing of a parthenogenetic strain to produce males, but since wFol induces parthenogenesis in addition to promoting host egg development, this experiment is not possible in the diplo-diploid F. candida. In several Hymenopteran species, Wolbachia has been shown to induce diploidy and feminization resulting in parthenogenetic reproduction [31,32] and several draft genomes of PI-Wolbachia from Hymenopteran host are available: wTpre from Trichogramma pretiosum and wUni from Muscidifurax uniraptor [33][34][35]. Building a larger database of Wolbachia genomes associated with PI might shed more light on the genes involved in PI, similar to recent finding for CI [36].
In this study, we set out to explain the factors that contribute to the expansion in the wFol genome, which can help further understand the genomic evolution within endosymbionts. Therefore, we updated and re-annotated the wFol genome and compared it to a diverse set of high quality Wolbachia genomes. In addition, we searched for genes shared by Wolbachia associated with PI that could aid in understanding the mechanism behind this manipulation. To this end, we assembled a draft genome of an additional PI strain, wLcla, that has been shown to cause diploidy restoration in the parasitoid wasp Leptopilina clavipes [31]. Finding the genes underlying PI in Wolbachia might elucidate the mechanism behind it and could resolve the debate on the nature of wFol host interaction.

Genome assemblies of wFol and wLcla
The short read corrected wFol genome consisted of one contig with a total length of 1,801,626 bp and a GC-content of 34.35%. This was 43 bp longer than the initial assembly and the GC-content was a half percent higher [24]. The draft genome of wLcla was assembled in 46 contigs with a total length of 1,150,755 bp and a GC content of 34.11%. Half of the assembly was contained in 9 contigs (L50) larger than 43,523 bp (N50). For a Wolbachia genome this is relatively small, but very similar in size to the other PI strain from supergroup B (wTpre) [34]. However, wTpre is in one scaffold that contains a total gap length of 16,680 bp and wLcla is in 46 pieces, thus we still miss information for both of these genomes. Hence, the current assembly of wLcla probably represents the major part the complete genome, but it will presumably increase slightly in size upon completion.

Annotation of wFol and wLcla
To evaluate the quality and completeness of our annotations, we run the BUSCO pipeline based on 148 essential bacterial genes on the annotations of wFol, wLcla and an additional 16 Wolbachia annotations to compare with (see Additional file 1). The wFol genome contained the most complete set of essential genes of the 18 Wolbachia analysed. It contained the complete sequences of 92.6% of the essential genes, including two duplications, and two fragmented. Only nine genes were missing, which was the lowest number of all Wolbachia analysed. The BUSCO gene set to which it was compared is a curated set of genes that are essential to most bacteria. However, in Wolbachia even complete genomes miss 9 to 25 genes from the BUSCO set, indicating that these genes probably are not missing from the assemblies and annotations but because they have become redundant for Wolbachia due to its endosymbiotic lifestyle. In the draft genome of wLcla we found 133 complete (89.8%), two fragmented and 11 missing genes, which is very similar to most of the other Wolbachia genomes.
Annotating the improved wFol genome uncovered a total of 1472 protein coding sequences (CDS), which is the largest number of CDS for any Wolbachia genome sequenced so far. This is not surprising given that it is the largest Wolbachia genome, and genome size and gene number correlate well in bacteria. However, a remarkably low number of pseudogenes were found in its genome. Only 2.8% of the genes appeared to be pseudogenized, while the pseudogene content in the other Wolbachia genomes ranged from 4.7 to 22.1%. In addition, we observed a high number of transposases, DNA-repair related genes and ankyrin repeat containing proteins (ANKs; see Table 1). The number of ANKs (96) and DNA-repair related genes (34) in wFol outnumbered the quantity found in any other Wolbachia and the number of transposases is second only to the number found in wCle.
In addition, we found large numbers of phage genes, which were concentrated in five regions of phage origin (RPO) that were labelled WOFol1 to 3 and WO-like island 1 and 2 (see Additional file 2: Table S1). With a combined size of 458,452 base pairs, they took up 25.4% of the total genome size of wFol, the largest amount in any Wolbachia. Remarkably, pseudogenes and the three before-mentioned overrepresented groups of genes (ANKs, DNA-repair related and transposases) were all unequally distributed between the phage regions and rest of the genome (see Fig. 1a). Of all groups more than the expected 25.4% was present within the RPOs, finding 29 out of 44 pseudogenes (binomial-test, R, p < 0.001), 23 out of 34 DNA-repair related genes (binomial-test, R, Column label abbreviations: CDS stands for coding sequences and ANK for ankyrin repeat containing protein. Annotations of wCle, wTpre and wUni did not specify pseudogenes; for these strains this category is therefore labelled as not available (NA) p < 0.001), and 42 out of 96 ANKs within them (binomial-test, R, p < 0.001). The distribution of transposases over the phage regions and rest of the genome was not different from a random expectation, since 26.6% of them were found in the RPOs (binomial-test, R, p = 0.75). However, when focussing on individual transposase families, they did show unequal distribution patterns (see Fig. 1b). While the IS4, IS110 and IS5 families were proportionally distributed over the whole genome, the PD-(D/E)XK family was more common in the RPOs than expected (binomial-test, R, p < 0.001). In addition, two of the other five families exclusive to either the RPOs or to the chromosomal part of the genome also showed a significantly skewed distribution (IS481: binomial-test, R, p < 0.01; IS5, IS903: binomial-test, R, p < 0.05). The annotation of wLcla uncovered 880 CDS and 194 pseudogenes (see Table 1). Thus, while it is similar in size to wTpre, it contains 525 fewer CDS and 194 more pseudogenes. This is probably an effect of differences in annotation style, as no pseudogenes were annotated for wTpre, but many truncated genes were found [34] that were therefore counted as one or two CDS instead of one pseudogene. Finally, wLcla contained 4 phage-related genes as well as 7 phage-related pseudogenes, thus showing signatures of remnant phages. However, no conserved RPOs could be detected within the assembled contigs.

Phage annotation and synteny
As the RPOs take up more than a quarter of the wFol genome we put extra effort in their annotation. Kent et al. (2011) compared several WO phages and found that they are composed of multiple modules of genes that link to certain functions. We also found such modules in the WOFol phages, which included patatin-like phospholipases and baseplate, head and tail modules (see  Table S1). Surprisingly, all three WOFol regions contained the tail module, which is often lost in other WO [37]. In addition, WOFol1 was missing the head and baseplate modules, but did contain the conserved WD0611-WD0621 cluster that is also found in several other WO phages [38]. All three WOFol regions also contained the recently described eukaryotic association module (EAM) [38] and within them 6 of the WO Latrotoxin CTD proteins. Further, the RPO of wFol contained 19 resolvases, which is at least two times the number of resolvases found in any other Wolbachia genome (see Table 1). Resolvases are often connected to phage integration at specific sites, while integration by transposases is less site-specific [39]. Finally, WOFol 2 and 3 also harbored a bacteriophage abortive infection system, the AbiEi and AbiEii antitoxin-toxin complex, which has never been found in WO before. This system can provide phage protection at the population level by killing its host when infected by a new phage [40].
The large number of RPOs in wFol led to the question if all these regions arose from a single ancestral phage that duplicated itself or, whether multiple phages infected the wFol genome. Therefore, we searched for collinear blocks in the RPO and found them both between A B Fig. 1 Distribution of genomic elements of interest in the wFol genome. Legend: a) Circular map of wFol displaying the distribution of genes of interest between regions of phage origin (RPOs) and chromosomal parts of the genome. Each ring gives locations of certain genes or genomic elements. From outer to inner ring: CDS forward strand, CDS reverse strand, RPO, transposases, ankyrin repeat containing genes (ANKs), pseudogenes and DNA-repair related genes. RPOs are separately labelled: WO regions with WOFol1 to 3 and WO-like islands with WOI1 and 2. b) Double side histogram that presents the distribution of transposase groups between the chromosomal and the RPO. Distributions were analysed with a binomial test in R, asterisks correlate to significant levels * to p > 0.05, ** to p > 0.01 and *** to p > 0.001 the three WOFol regions and within WOFol2 and 3 (see Fig. 2a, b). However, based on the mosaic structure of the collinear blocks there does not seem to be a complete duplication of one of the WO phages.

Orthologue identification and orthogroup expansion
Orthofinder was run to cluster the complete set of 19,303 protein-coding sequences (CDS) of the 18 Wolbachia into orthogroups resulting in 1239 orthogroups containing 18,480 CDS (95.7%) (see Additional file 4: sheet Overall statistics). Of these, 460 were present in all Wolbachia genomes, including 414 that only contained single copy genes (see Additional file 4: sheet Orthogroups shared by all). These 460 genes can be considered the Wolbachia core genome, which is within the range of the 489 orthogroups found by Brown et al. (2016), when comparing members of supergroups A-D, F and L. For wFol, 1346 of the 1472 CDS were grouped into 936 of the orthogroups (see Additional file 4: sheet All orthogroups). Thus, 126 CDS remained unassigned and were therefore considered to be unique to wFol (see Additional file 4: sheet Unassigned genes). wTpre was the only genome containing more unique CDS. This was most likely due to the many truncated/pseudogenes in this genome annotation [34], which probably have ended up in separate groups because of Orthofinder's algorithm sensitivity for gene length [41], thus creating false orthogroups of truncated genes. Most of the unassigned genes from wFol were hypothetical proteins (76) and putative membrane proteins (27), but 23 of these genes could be annotated in more detail (see Additional file 4: sheet Unassigned genes). Moreover, 35 (27.8%) of the unassigned CDS were located within the RPOs, which include the AbiEi and AbiEii antitoxin-toxin complex mentioned above (Additional file 5). For the wLcla draft genome, 875 of their 879 CDS were grouped into 817 orthogroups, while only 4 remained unassigned. Three of these were annotated as hypothetical proteins and one was a transposase (see Additional file 4: sheet Unassigned genes). Therefore, no distinct functions could be linked to these genes.
To define orthogroup expansions and contractions, Z-scores were calculated for each orthogroup (see Fig. 3 and Additional file 4: sheet all orthogroups). Z-scores measure the deviation of the number of genes of one strain from the average number of genes per strain in an orthogroup. Orthogroup expansions had more genes (Z-score 2 or higher) and contractions fewer (Z-score − 2 or lower).
The three genomes with the highest number of gene family expansions were all linked to PI (wTpre, wUni and wFol). Unfortunately, no orthogroup was expanded in all three Wolbachia genomes. Focussing on the wFol genome we found 72 expanded and 10 contracted orthogroups (see Additional file 4: sheet Expanded or contracted in wFol). All 10 contracted orthogroups were genes completely missing in wFol. The 72 expanded orthogroups contained 371 genes of which more than half within the RPOs (194 genes within 21 orthogroups). In line with this, 89 genes were phage-related genes, pressing the notion that the expansion of wFol is for a large part due to an increase of phage genes. In addition, more than half (190) of the genes were transposases (93), ANKs (74) and DNA-repair related (21), proving that these groups are overrepresented in this genome.
Not surprisingly, considering its small size the wLcla genome contained only 14 expanded orthogroups with 26 genes in total (see Additional file 4: sheet Expanded or contracted in wLcla). Eight of these genes were within one orthogroup of IS110 family transposases, while all other groups consisted of 1 gene missing in all other Wolbachia or 2 duplicate genes in wLcla. Most of these other groups consisted of uncharacterized proteins without a clear function and the genes with annotation had diverse functions in several biological processes. The wLcla genome contained only five significantly contracted orthogroups, which is surprising given the large number of pseudogenes. This indicates that many of the pseudogenized genes of wLcla were also degraded in other Wolbachia.

Parthenogenesis induction genes
Clustering all genes into orthogroups also allowed us to search for orthogroups whose absence or presence was associated with the four PI-Wolbachia strains (wFol, wLcla, wTpre and wUni). We found no genes that were exclusively missing from the four PI-Wolbachia. There was a large number of shared orthogroups (684) between the four genomes, which obviously also included the 460 core genome orthogroups (see Fig. 4a). However, there were no orthogroups that were exclusively present in all four or three out of the four parthenogenesis-associated Wolbachia genomes (see Fig. 4b). Only between pairs of the PI-Wolbachia 11 exclusive orthogroups existed (see Additional file 4: sheet Exclusive or mainly in PI). Five of those contained only uncharacterised proteins and another two contained only transposases. The remaining four consisted of: 1) a putative phage protein shared by wUni and wFol, 2) a protein with a magnesium transported domain shared between wTpre and wUni and 3) a cluster of two genes shared between wTpre and wLcla consisted of a protein with a partial fungal domain of unknown function and a protein that contains several leucine-rich repeats. These leucine-rich repeats are protein binding domains that are involved in a wide range of biological processes [42]. Furthermore, we also searched for orthogroups that were present in at least three of the four PI strains and in a maximum of two other Wolbachia strains. This yielded two orthogroups with single copy uncharacterized genes, represented in wFol by wFol_04740 and wFol_12640, and both shared by wFol, wLcla, and wTpre (see Additional file 4: sheet Exclusive or mainly in PI). Absence of these genes in wUni strain might be inaccurate, as the still quite fragmented assembly is possibly hampering gene annotation (Table 1). One orthogroup was also found in the male-killing strain wBolb-1 and the other orthogroup in both wBolb-1 and CI-strain wPip_Pel.
Based on their annotation neither of the examined genes could be directly linked to parthenogenesis induction.

Whole genome correlations
We found a positive correlation between Wolbachia genome size and the amount of phage DNA it contained (see Fig. 5a). This indicates that a significant portion of the variation in genome size was accounted for by the RPOs. The genomes of Wolbachia strains that are essential to their host occurred at both ends of the size spectrum, with most having small genomes and little phage DNA and wFol having a large, phage-rich genome. In addition, we also found a positive correlation between the number of DNA-repair related genes and the genome size of Wolbachia (see Fig. 5b).
Host genome size and gene number were also correlated to Wolbachia genome size and gene number, respectively (see Fig. 5c, d), with gene content being more strongly correlated. Moreover, the correlation between genome sizes still holds when RPO are excluded (ρ = 0.86, P = 0.01; see Additional file 6), although the deviating strain wCle needed to be excluded as the RPO could not be properly annotated and therefore their size could not be determined. These correlations suggest that the genome size and number of genes of a Wolbachia are not independent from the size and number of genes of its host genome.

Discussion
The wFol genome is the largest Wolbachia genome to date, and our results suggest that its genome size has increased due to the integration of several phages, as phage regions take up more than a quarter of the total wFol genome. This is supported by an expansion of 21 orthogroups containing phage genes. These regions of phage origin (RPOs) had higher numbers of ankyrin repeat containing genes (ANKs), DNA-repair related genes, pseudogenes and PD-(D/E)XK transposases. In addition, they contained a bacteriophage abortive infection system new to Wolbachia. Comparing wFol, wLcla and two other PI-associated strains to a diverse set of 14 Wolbachia genomes did not elucidate a set of genes unique to the four PI strains. However, there were unique sets of genes between some pairs of the PI associated strains. Finally, we found strong correlations between Wolbachia and host gene numbers, suggesting that there might be convergent evolution between Wolbachia and their host genomes.

Genome expansion in wFol
The first question we set out to answer was why the genome of the obligate Wolbachia endosymbiont of the parthenogenetic springtail F. candida (wFol) has expanded, while current theory predicts an obligate relationship to lead to a reduced genome size [12,18]. Obligate endosymbionts usually have a small genome size and contain few repeated and mobile elements such as transposases [43], ANKs [44,45] and RPOs [46,47]. Genome reduction of this kind is also evident in the genomes of obligate Wolbachia strains in filarial nematodes (wOo, wOv and wBm), which are the smallest complete Wolbachia genomes, containing no or very few transposable elements and phage derived genes [9,10,48]. In contrast, the genome of the obligate wFol strain is the largest complete Wolbachia genome discovered to date with a length of 1,801,626 base pairs (bp) A B Fig. 4 Orthogroups shared by PI Wolbachia. Legend: Venn diagrams representing a) a distribution of all orthogroups shared between the PI associated Wolbachia, including the orthogroups shared with the other 14 Wolbachia and b) a distribution of all orthogroups exclusively present in the PI associated but not in the other 14 Wolbachia [24]. Moreover, upon annotating this unusually large genome meticulously, we found that it is highly enriched in repeated and mobile elements and that the RPOs take up more than a quarter of its genome. These regions contain many of the ANKs and transposases that are enriching this genome and have a big influence on the genomic structure and size of wFol. Many expanded orthogroups were also found to be located in the RPO, including a set of genes connected to DNA repair. Integrated phage genes within endosymbionts can still be expressed and are known to influence bacterial and host processes [36,49]. Therefore, the functions of these integrated genes may affect the performance of the tri-partite symbiosis. Typically, in obligate endosymbiont genomes a depletion of DNA repair genes is seen and the loss of these genes would result in a higher effective mutation rate [13]. Thus, the gain in DNA repair genes in the wFol genome can be expected to lower effective mutation rate, resulting in a better-conserved genome with fewer deletions and fewer pseudogenized genes.
We also found a large number of unique genes present in wFol. Although RPOs are known to be a source of new genes [38,45,50], in wFol unique genes were evenly distributed over the genome. A possible explanation for the large number of novel genes could be that wFol is the first annotated Wolbachia genome of supergroup E, while many of the other supergroups were represented with more genomes in our analysis. Therefore, this group of unique genes might not just represent genes unique to wFol but also include genes that are specific to the entire E supergroup. Nevertheless, the presence of this many unique genes indicates that either wFol specifically or all members of supergroup E are able to acquire new genes or preserve existing ones more easily compared to other Wolbachia.
Another strong signature of gene conservation is the low number of pseudogenes found in wFol. pseudogenes that were present were located within the RPO, suggesting a less stringent conservation of phagederived genes. This is in line with what is seen in other Wolbachia where the phage regions are often more prone to degradation [51]. However, the rather high share of intact phage genes within wFol, suggest that even these regions with higher degradation rates are still being maintained at an elevated rate. The combined results on phage preservation, the low number of pseudogenes, the many DNA repair genes, and the large number of unique genes consistently point towards a genome shaped by gene preservation. This is inconsistent with expectations based on the symbiont genome reduction model. At first glance, the wFol genome with its relatively large size and many mobile elements resembles a recently host-restricted symbiont. However, these symbionts are generally facultative for their host, harbour many pseudogenes and suffer large and small deletions, as their genomes change rapidly [12,18]. By contrast, wFol is obligate to its host, possesses only few pseudogenes, and shows no signs of large deletions. Furthermore, the start of the symbiosis between wFol and its host probably coincided with the host becoming parthenogenetic. Although, this split has not been properly dated yet, genetic evidence shows considerable differentiation between sexual and parthenogenetic lines of Folsomia candida [29], suggesting this was not a recent event. Thus, wFol does not fit the description of a reduced long-term obligate symbiont genome, nor a recently host-restricted symbiont genome. Rather, it seems to fall into a new category of a long-term obligate symbiont that was able to avoid genome degradation.
This leads to the question: what has caused the genome preservation in wFol compared to the other Wolbachia, including the other PI and obligate strains? These obligate strains are just like wFol vertically transmitted within hosts that are dependent on their Wolbachia, thus there must be another explanation for the differences in genome maintenance than level of interdependence with their host or form of transmission. Interestingly, there are unique features to the genome of F. candida, which mirror the genomic patterns of its symbiont. Its genome has with 28,734 genes the most gene rich genome of the Wolbachia hosts (Fig. 5d). Moreover, it is also gene richer than the two other published Collembolan genomes of Holacanthella duospinosa where they found 12,000 gene models [52] and Orchesella cincta with 20,249 genes [53]. In addition, groups of genes related to DNA repair expanded in both genomes, with F. candida having expanded groups of ATP-dependent DNA helicases which are important for DNA repair [24,54]. Finally, both host and endosymbiont contain high amounts of foreign DNA, as 2.8% of F. candida genes are horizontally transferred genes (HTG) from a wide variety of organisms including bacteria, fungi and protists, but not from Wolbachia. This might sound low compared to the 25.4% of phage DNA in wFol but is in fact the highest percentage found in any metazoan genome except for rotifers and some nematode species [24] Moreover, this abundance of mostly intact HTG was correlated with an increase in transposons. Thus, also the F. candida genome seems focused on preserving genes. This is as far as we know, the first case that shows signs of convergent evolution between endosymbiont and host genomes. In addition, the correlation between Wolbachia gene number and host gene number of all combinations with usable genomes, suggest that convergence between host and endosymbiont genomes may have taken place in more Wolbachia strains.

Regions of phage origin
The integration of phage DNA is one of the main reasons for the large size of the wFol genome. The RPO included three phage WO regions and two phage WO-like islands, which is within the range of two to five prophage WO haplotypes found in other fully sequenced Wolbachia genomes [37]. However, the size of the wFol WO regions is much larger and two of the regions contain multiple copies of essential phage gene clusters. Possibly, multiple phages clumped together within the wFol genome or the phages that integrated had multiple copies of the same modules. Alternatively, a recent duplication took place but this is not very likely because the collinear blocks between phages were very fragmented and the longer blocks contained mainly the conserved modules found in all WO phages. Such large clusters containing multiple sets of the same phage modules have previously been found in Wolbachia genomes wPip and wNo [50,55] and hamper an assessment of the exact number of phages that integrated within this genome.
All three WO regions within wFol contained the characteristic elements and standard modules of phage WO. These included the Patatin gene that is probably helpful for entering the host cell [37] and the recently defined eukaryotic association module including latrotoxin-CTD domain proteins, which are related to black widow venom genes [38]. Neither of the WO regions contained the cifA and cifB genes linked to CI [36], therefore it is very unlikely that wFol can cause CI. Some wFol RPO features are exceptional, such as the vast amount of resolvases and transposases, which are thought to be involved with phage integration into their bacterial host [39,56]. However, integration by transposases usually takes place at non-specific integration sites while resolvases use conserved sites [39]. Thus, the excessive presence of both transposases and resolvases in the wFol phages raises questions concerning their mode of integration. Finally, a toxin-antitoxin system that is new for Wolbachia was found in the WOFol2 and 3. This AbiEi and AbiEii system is an abortive infection system that can cause altruistic cell death to prevent new phages from settling within the bacterial population [40]. One might argue that such a system may prevent further genome expansions by additional phages.

Genetic basis of parthenogenesis induction
We also perused the wFol genome for genes that are associated with PI to elucidate the genetic basis of this reproductive manipulation. Together with three other PI Wolbachia, wUni of supergroup A and wTpre and wLcla of supergroup B, they were compared to a diverse set of 14 Wolbachia including members of 6 different supergroups (A,B,C,D,F and L). We searched for gene sets present in at least three PI and a maximum of two other Wolbachia strains, comparable to the CI phenotype where a toxin-antitoxin gene cluster was shared by all CI inducing or repressing strains and some others [36,57]. We did not find any genes present in all PI strains, which may have several non-mutually exclusive explanations. First, PI genes may not be uniquely present in Wolbachia expressing the PI phenotype, but also in some of the Wolbachia causing other reproductive manipulations, similar to what is seen for the CI phenotype [58]. The two genes we identified to be associated with three of the PI-strains were also present in wBolb-1 and in wPip_Pel that are both in supergroup B. The presence of these genes in the two other strains might have to do with the fact that phenotypic expression of Wolbachia has been found to be determined in combination with the host genotype. For example, the CI-strain wRec caused male killing after being transferred to a new host [59] and wTei caused CI after a host transfer while showing no signs of reproductive manipulation in its natural host [60]. This same interaction with host genotype may apply to PI expression. A second possibility is that the PI phenotype is not a single genotype but is achieved via different cellular mechanisms formed by convergent evolution, in which case we do not expect a single gene cluster to be shared by all PI-Wolbachia. Support for this explanation can be found in studies showing that parthenogenesis is induced in a two-step mechanism of diploidization of the gamete followed by feminization of the egg [61,62]. Diploidization can be achieved in different ways. In both Trichogramma wasps and Leptopilina clavipes (the hosts of respectively wTpre and wLcla, the two PI inducing Wolbachia of supergroup B) gametes are rendered diploid by failure of the first mitotic anaphase division [31,63]. In Muscidifurax uniraptor (host of wUni of supergroup A) diploidization is only restored after the second mitotic division through a fusion of the two mitotic nuclei [64]. For, wFol of supergroup E the speculated mechanism is diploidy restoration through either non-disjunction or fusion of the two haploid division products at the end of the first anaphase [65] and thus similar to the situation in wTpre and wLcla, although the difference in sex determination system might call for a different manipulation for the same outcome. Therefore it has already been suggested that diploidy restoration in Wolbachia most likely has a polyphyletic origin [31]. With this in mind, the cluster of two genes uniquely shared by wTpre and wLcla might still be very interesting and might be linked to one of the two steps of PI in these lines that seem to have a similar mechanism developed in supergroup B.

Conclusions
Large RPOs with ample repair genes and accumulation of repetitive and transposable elements make up most of the expansion of the wFol genome. This genomic signature of gene conservation was mirrored in the F. candida host genome. We found that a large part of the variation within the genome size and gene number of facultative endosymbionts is correlated to the gene number of the host. This suggests that host and symbiont genome may be subject to correlated selection pressures that resulted in convergent evolution between host and endosymbiont, or that somehow the host may have a direct influence on the symbiont genomes. However, these selection pressures would probably be neutralized after endosymbionts move into a specialized cell, explaining the well-documented steady genome decrease seen in those cases [11,66,67]. Therefore, genome reduction may ultimately result from becoming obligate. However, before the endosymbiont resides in a specialized cell, but while being mainly transmitted vertically, genome expansion of endosymbiont genomes might occur.
The search for the PI genes yielded a set of potential candidate genes. Elaborating on these findings could answer whether the PI genes are monophyletic or polyphyletic. The complexity of this trait and the indications that this is caused by multiple genes could also means that both are not mutually exclusive in this case. . Illumina reads were mapped to the assembly with BWA [68] using default parameters. Variants between the assembly and the mapped Illumina reads were called and quality filtered using GATK v. 3.7 [69], filtered based on read depth with vcffilter ("DP > 10") of the Vcflib package (E. Garrison, 2012, https://github.com/vcflib/vcflib). Variants were inspected manually and when they were supported by the mapped Illumina data they were included in the genomic sequence.

Assembling of wFol and wLcla
Reads used for the wLcla assembly were taken from a study that sequenced its host Leptopilina clavipes [70] using both an Illumina HiSeq (HiSeq) and a Pacific Biosciences RS I (PacBio) library. Hiseq reads were used to error correct the PacBio reads with the PacBioToCA pipeline of CeleraAssembler7.0 [71]. We used the corrected PacBio reads for the wLcla assembly. To filter out the Wolbachia reads, all PacBio reads were mapped onto the wTpre and the wPip_Pel genome with BLASR [72]. A consensus based on the mapped reads was made per reference genome with PBDAG-Con [73] and the resulting contigs were extended and joined with PBjelly [74]. Thereafter, all reads were mapped back to these preliminary assemblies and PBDAG-Con and PBjelly were run again to extend the preliminary contigs further. All reads mapping to these two assemblies were extracted and assembled de novo in Mira [75] with the following parameters: COMMON_SETTINGS -SK:mmhr = 1 PCBIOHQ_SET-TINGS -CO:mrpg = 5. Next, all corrected Pacbio reads were mapped back to the Mira assembly, after which two iterations of the combination of PBDAG-Con and PBjelly were run to extend and connect the contigs as much as possible. The process ended with a final step of PBDAG-Con to confirm the extensions based on mapped reads. Finally, the HiSeq data was mapped to the assembly, variants were called, manually checked and the assembly was adjusted in the same way as described for the wFol assembly.

Annotation of wFol and wLcla
The annotation of both genomes was done using the DIYA pipeline [76], in which we included: Prodigal for gene prediction [77], tRNAscan-SE and RNAmmer to predict RNA features [78,79] and GenePRIMP to mark possible pseudogenes and short genes without annotation [80]. Predicted genes smaller than 100 amino acids without hits in any of the databases were removed. Genes were manually annotated as pseudogenes when they contained frameshift mutations, premature stop-codons, missing start codons or disruptions by IS-insertions. In addition, protein domains were predicted according to the Pfam database [81], BlastP to the NCBI database [82] and FASTA searches [83] against a well-curated in-house database by Lisa Klasson of Wolbachia genomes. All results were loaded into Artemis [84] in which they were manually curated. Transposases were blasted to the IS-finder database to determine the family and group (https://www-is.biotoul.fr). Uncharacterized/unique proteins were run through the InterPro databases with InterProScan [85].
Genes that did not exhibit features indicative of any specific function, but did contain transmembrane segments, cytoplasmic and non-cytoplasmic domains were characterised as putative membrane proteins.
For wFol, equal distribution of gene groups and transposase families over chromosomal parts (74.6%) and RPOs (25.4%) was tested with binomial tests in R [86].

Annotation completeness and ortholog identification
We compared the wLcla and wFol assemblies to a set of 16 other Wolbachia genomes, selected based on assembly quality as well as phylogenetic and functional diversity (Additional file 2: Table S2) [9, 10, 28, 33-35, 50, 51, 55, 87-92]. Protein sequences of these 16 genomes were downloaded from the NCBI database and the completeness of their gene content was predicted with the BUSCO v3 pipeline, which compares the Wolbachia genomes to a set of 148 single copy bacterial genes (Bacteria odb9) [93]. Furthermore, orthogroups were inferred using OrthoFinder and are defined as groups of genes that all derived from a single gene in the last common ancestor [41]. This allowed us to look for orthogroups shared between Wolbachia associated with PI and find unique genes (genes lacking orthologues in other Wolbachia).
In all genomes, expanded and contracted orthogroups were identified by calculating the z-scores [94], which is calculated by subtracting the average number of genes in an orthogroup over all species from the gene number of a focal species and dividing this by the standard deviation. A z-score of 2 or above indicates an expansion and a score of − 2 or below a contraction of an orthogroup.

Prophage annotation and synteny
Prophage regions of wFol and wLcla were initially identified by PHASTER [95]. While this correctly finds the more common phage genes it does not identify the more diverse EAM [38], as they contain of eukaryotic genes picked up by phages that are therefore not being recognised as phage genes. Therefore, the EAM were defined during manual curating of the prophage regions. Phage region lengths of other phages were based on annotations from  to include the whole phages with EAM or, if not available from that paper, by determining them in a similar way (Additional file 2: Table S1). As, the phage region are often the most difficult parts to assemble, this was only done for Wolbachia assemblies with two or less scaffolds, to avoid the use of incomplete sets of RPOs within our analysis. Moreover, wCle could not be used as phage genes seemed to be removed from its annotation.
Collinearity between the regions of phage origin (RPOs) was analysed with the MCScanX package [96]. A BlastP search of all against all RPO protein sequences was performed with an E-value cutoff of 1e− 10, followed by a MCScanX run (−m 2). Synteny plots were drawn using Circos [97].

Whole genome correlations
To examine overall genomic expansion and contraction patterns and their possible connection to phages and their host, we assessed correlations between Wolbachia genome size versus amount of phage DNA, Wolbachia genome size versus host genome size, and number of Wolbachia genes versus number of host genes for all combinations that had sufficient reliable data. Genomic information of all available hosts of the Wolbachia analysed in this study was collected from online servers, most were downloaded from NCBI (hosts and accession numbers: Brugia malayi, GCA_000002995.4; Cimex lectularius, GCF_000648675.1; Culex quinquefasciatus, GCA_000209185.1; Drosophila melanogaster, GCF_00000121 5.4; Drosophila simulans, GCA_000754195.2; Folsomia candida, GCA_002217175.1; Onchocerca ochengi, GCA_0009 50515.2 and Trichogramma pretiosum, GCF_000599845.1) and data for Onchocerca volvulus strain Cameroon was collected from http://parasite.wormbase.org/Onchocerca_vol vulus_prjeb513/Info/Index/. Spearman' s rank correlations were calculated in R.