Until the past decade, the genetic basis of CI remained elusive. The intangibility of the CI genes was due in no small part to the inability to genetically engineer symbionts such as Wolbachia (Iturbe-Ormaetxe et al., 2007; Thiem, 2014). Progress in CI genetics became possible with the genome sequencing of the wMel Wolbachia of D. melanogaster in 2004 (Wu et al., 2004). wMel has a streamlined genome with numerous mobile elements including phages (Wu et al., 2004). Notable among these elements was Wolbachia’s phage WO, which was first identified in 1978 via transmission electron microcopy of viral-like particles in symbiont-bearing Cu. pipiens (Wright et al., 1978). Prophage WO encodes a set of proteins termed the Eukaryotic Association Module that share homology to eukaryotic proteins, likely due to lateral gene transfer from eukaryotes to the phage, and they are predicted to interact with eukaryotic processes (Bordenstein and Bordenstein, 2016). wMel’s genome is also enriched with ankyrin proteins that are involved in protein-protein interactions in eukaryotes and are relatively rare in free living bacteria as compared to intracellular bacteria and eukaryotes (Al-Khodor et al., 2010; Jernigan and Bordenstein, 2014). Conversely, genome sequencing of the mutualistic wBm Wolbachia of Brugia malayia nematodes revealed it did not contain phage WO nor an enrichment of ankyrins (Foster et al., 2005). These findings suggested a correlation between reproductive parasitism and the presence of phage WO and/or ankyrin genes and motivated hypotheses that phage WO may be involved in CI (Foster et al., 2005; Wu et al., 2004; Yamada et al., 2011). Numerous additional genome sequencing projects would also be integral to identifying candidate genes for CI and rescue, including wPip, wAu, wRi, wHa, and wRec (Klasson et al., 2009; Klasson et al., 2008; Metcalf et al., 2014; Salzberg et al., 2009; Sutton et al., 2014).

The first attempt to functionally dissect CI’s genetic basis generated a list of 12 gene candidates in the wMel genome based on putative host interaction: nine ankyrin genes (WD0294, WD0385, WD0498, WD0514, WD0550, WD0633, WD0636, WD0754, and WD0776), two virulence-related genes (WD0579 and WD0580), and one phage-associated methylase gene (WD0594) (Yamada et al., 2011). Since Wolbachia are not genetically tractable (Iturbe-Ormaetxe et al., 2007; Thiem, 2014), D. melanogaster transgenic tools were used to test these gene candidates (Duffy, 2002). However, transgenic expression of these genes in aposymbiotic male flies revealed that none recapitulated CI (Yamada et al., 2011). Moreover, neither transcriptional nor genetic variation of Wolbachia’s ankyrin genes correlated with a strain’s ability to induce CI (Duron et al., 2007b; Papafotiou et al., 2011).

Additional ‘omic studies would pave the way for identification of new gene candidates. First, mass spectrometry and SDS-page analyses of spermatheca (the female sperm storage organ) extracts from symbiont-bearing Cu. pipiens females revealed the prophage WO protein WPIP0282 (Beckmann and Fallon, 2013), thus elevating a new candidate for CI and/or rescue and providing additional support to hypotheses that phage WO genes may contribute to CI. Second, genomic comparisons of the wMel genome against the genome of the non-parasitic wAu strain of D. simulans revealed nine genes absent in the non-parasitic strain that were present in wMel (Sutton et al., 2014). These candidates included numerous genes from Wolbachia’s prophage WO including WD0631 that is a wMel homolog of wPip’s WPIP0282, WD0632 which is adjacent to WD0631, and a set of transcriptional regulators (Sutton et al., 2014). Thus, evidence continued to build around phage WO genes as CI factors. Finally, sequencing of the wRec genome revealed a highly reduced prophage WO with approximately one-quarter the number of genes in a close relative (Metcalf et al., 2014). These genes contained several previously described candidates including the wRec homolog of WPIP0282, four transcriptional regulators also absent in wAu, and a Wolbachia transcriptional regulator gene wtrM linked to regulation of host meiosis in Cu. molestus (Beckmann and Fallon, 2013; Pinto et al., 2013; Sutton et al., 2014).

The reduced wRec genome would later form the basis of an unbiased, comparative ‘omic study assessing the genomes of CI-inducing Wolbachia, a genome of a non-parasitic strain and a transcriptome and proteome of Wolbachia-carrying ovaries (LePage et al., 2017). This analysis revealed only two phage WO genes in the Eukaryotic Association Module, WD0631 and the adjacent WD0632, as CI candidate genes in the wMel strain of D. melanogaster (LePage et al., 2017). These genes would also later be determined to be absent in the parthenogenesis-inducing wTpre strain of Trichogramma wasps (Lindsey et al., 2018). WD0631 and WD0632 were named cytoplasmic incompatibility factors A and B (cifA and cifB), respectively (LePage et al., 2017). The gene is referred to in lowercase and italics (cifA and cifB), the protein is referred to in uppercase with no italics (CifA and CifB), and the strain that the specific cif gene comes from can be defined with the strain name as a subscript (cif_wMel or cif_wPip). This gene nomenclature is consistent with guidelines from the American Society for Microbiology ('The Journal of Bacteriology, 2018').

With independent ‘omic identification of cifA and cifB as candidates for CI (Beckmann and Fallon, 2013; LePage et al., 2017; Sutton et al., 2014), two studies simultaneously explored the relationship between cif_wMel (LePage et al., 2017) and cif_wPip (Beckmann et al., 2017) genes and CI using transgenic expression systems in D. melanogaster. Singly expressing cifA_wMel or cifB_wMel in aposymbiotic D. melanogaster males failed to induce CI, but dual expression of the genes caused rescuable CI-like hatch rates and cytological embryonic defects (LePage et al., 2017), suggesting that the cif_wMel genes cause CI only when expressed together. Similar results were reported when cifA_wPip and cifB_wPip were dually expressed in aposymbiotic D. melanogaster males, but rescue was not achieved (Beckmann et al., 2017), indicating that some biological or technical limitation of the system may have inhibited the ability to rescue transgenic cif_wPip CI in a heterologous expression system. Later, similar transgenic experiments revealed that cifA_wMel expression in aposymbiotic D. melanogaster females can rescue CI (Shropshire et al., 2018), motivating a Two-by-One genetic model of CI wherein cifA and cifB cause CI unless cifA is expressed in the ovaries or embryo to rescue it (Figure 3A). This model was further supported through transgenic expression of cifA_wMel and cifB_wMel in aposymbiotic males to induce transgenic CI and through crossing them to cifA_wMel-expressing aposymbiotic females to show that transgenic CI can be rescued at levels comparable to symbiont-bearing females (Shropshire and Bordenstein, 2019).

Notably, while the Two-by-One genetic model is most consistent with transgenic expression studies in D. melanogaster that achieve rescuable CI (Beckmann et al., 2017; Chen et al., 2019; LePage et al., 2017; Shropshire et al., 2018; Shropshire and Bordenstein, 2019), transgenic expression of cifB_wMel and cifB_wPip in yeast can cause temperature sensitive lethality that can be inhibited by co-expression with cognate cifA (Beckmann et al., 2017). Although since yeast do not have sperm or eggs, which are the targets of CI, the relevance of phenotypes observed in a heterologous yeast expression system need to be replicated in insect models. While a divergent cifB gene variant from wPip can weakly reduce embryonic hatching when crossed to aposymbiotic females (Chen et al., 2019), it remains unknown if this lethality can be rescued. These data lend support to the possibility that some strains may employ a model of CI wherein cifB is the CI-causing factor and cifA is only the rescue factor. However, interpretation of these results is significantly complicated by the absence of rescue data for any cifB-associated reduction in embryonic hatching in insects. Moreover, mutagenesis analyses, described in further detail below, indicate that changing conserved residues across CifA can crucially prevent CI, lending additional support for CifA’s important role as a CI-inducing factor (Shropshire et al., 2020). More functional genetic analyses will be necessary to provide evidence for an alternative to the Two-by-One genetic model of CI. Hereafter, we will discuss relevant phenomena in the context of a Two-by-One genetic model.

Initial comparative sequence analysis of Cif proteins revealed that CifA and CifB have concordant phylogenies with considerable divergence across several distinct phylogenetic clades (LePage et al., 2017). Since then, the availability of additional genomes and Cif sequences have exposed at least five clades referred to as Types 1–5 (Figure 3B; Bing et al., 2020b; Lindsey et al., 2018; Martinez et al., 2020), and highly divergent Cif-like homologs in Orientia and Rickettsia bacteria (Gillespie et al., 2018). It is likely that continued genomic sequencing will reveal additional Cif phylogenetic Types. The wMel Cif proteins belong to the Type 1 clade, and wPip has both Type 1 and Type 4 Cif proteins. The phylogenetic classification of a cif gene can be indicated with a T# subscript in brackets to the right of the gene name (i.e. cif_wMel[T1] or cif_wPip[T4]).

While the cif genes are associated within the Eukaryotic Association Module of prophage WO or WO-like islands (Bordenstein and Bordenstein, 2016), Cif phylogeny is not concordant with phage WO or Wolbachia phylogeny, potentially reflecting the typically high rates of inter- and intragenic recombination in phage genomes (Bordenstein and Wernegreen, 2004; LePage et al., 2017). Some cif genes are flanked by ISWpi1 transposons which may assist horizontal transfer between WO-associated regions or Wolbachia strains, but it remains unclear if they alone are responsible for divergence between the phylogeny of the cif genes, Wolbachia, and phage WO (Cooper et al., 2019; Madhav et al., 2020). To date, only cif genes belonging to the Type 1 and 4 clades have been experimentally evaluated and confirmed to cause and rescue CI (Beckmann et al., 2017; Chen et al., 2019; LePage et al., 2017; Shropshire et al., 2018; Shropshire and Bordenstein, 2019), but unpublished data suggest cif genes in the Type 2 clade can cause and rescue CI (JDS and SRB, unpublished data). These studies indicate that despite considerable sequence divergence, proteins across the phylogenetic landscape of the Cifs remain capable of causing and rescuing CI. Though, the phenotypic output of the Type 3, Type 5, and Orientia/Rickettsia cif-like genes have not been experimentally assessed, and it remains unknown if they can contribute to CI phenotypes. However, the CI-inducing strains wNo of D. simulans and wStri of La. striatellus only have Type 3 or Type 5 genes, respectively (Bing et al., 2020b; LePage et al., 2017), suggesting these genes may cause CI and rescue. Since CI has not been reported in Orientia or Rickettsia species, it is less likely these distant cif-like homologs contribute to CI (Gillespie et al., 2018). Alternatively, other yet identified gene sets may be CI-capable in CI-causing symbionts. Indeed, genomic and transcriptomic sequencing of CI-inducing Cardinium reveals that they do not carry obvious homologs to the CI genes (Mann et al., 2017). More functional genetic studies will be necessary to identify and assess the landscape of cif-like genes and to identify alternative CI genes.

Theory predicts that selection should favor the maintenance of rescue, but not CI, when Wolbachia are at equilibrium in the population since CI would no longer increase Wolbachia’s prevalence in the population (Turelli, 1994). In this context, and under the Two-by-One genetic model of CI in which cifA is involved in both CI and rescue while cifB is involved in CI, putative loss-of-function mutations in CI will accumulate preferentially in cifB, though not universally, relative to cifA. Consistent with this hypothesis, there are approximately twice as many putative loss-of-function mutations in cifB relative to cifA (Martinez et al., 2020), suggesting that CI is ablated more frequently than rescue. For example, the wPanMK and wPanCI Wolbachia strains of D. pandora cause male and CI respectively, but the male-killing strain has an early stop codon in CifB that putatively inhibits function and may allow for the phenotypic switch from CI to male (Asselin et al., 2018). Additionally, the wMau Wolbachia of D. mauritiana encodes Type 3 Cif proteins (LePage et al., 2017; Meany et al., 2019) and does not cause CI, but it can rescue CI caused by the closely related wNo of D. simulans (Bourtzis et al., 1998; Rousset and Solignac, 1995; Zabalou et al., 2008). CifA_wMau[T3] sequence is identical to the CI and rescue-capable wNo Wolbachia strain, but CifB_wMau[T3] has a frameshift that introduces over ten stop codons that in turn associates with the loss of CI (Meany et al., 2019). However, it is important to note that two CI-capable Wolbachia, wYak of D. yakuba and wRec of D. recens (Cooper et al., 2017; Shoemaker et al., 1999), have putative loss-of-function mutations in the form of truncations in all of their cifB genes (Martinez et al., 2020). These results suggest that while the homologs appear pseudogene-like, they may be functional. Indeed, dual expression of cifA;B_wRec[T1] transgenes in aposymbiotic males yields a rescuable hatch rate reduction (JDS and SRB, unpublished data). In sum, loss-of-function mutations are common in cifB, resulting in ablation of CI while maintaining rescue. Moreover, evolution-guided mutagenesis assays across the Cif proteins reveal that conserved residues in the CifA C-terminal domain of unknown function are crucial for CI (Shropshire et al., 2020). Thus, it is plausible that seemingly innocuous amino acid changes within this CifA region may ablate CI and yet maintain rescue. Higher resolution comparative genomic analyses coupled with phenotypic data will be necessary to evaluate this hypothesis. More work will also be necessary to functionally assess the impacts of putative loss-of-function mutations on CI and rescue capabilities.

While the above examples clearly highlight a relationship between cif sequence variation and loss of CI, there are other strains that are more difficult to explain. For example, wSuz of D. suzukii encodes both Type 1 and Type 2 cif genes that are highly similar to the strong CI-inducing strain of wRi, but wSuz does not cause CI (Cattel et al., 2018; Conner et al., 2017; Hamm et al., 2014; Lindsey et al., 2018). While CifA_wSuz[T2] has been disrupted by the insertion of a transposase, the Cif_wSuz[T1] gene pair remains intact and has only 2–4 amino acid substitutions relative to Cif_wRi[T1] (Conner et al., 2017; Lindsey et al., 2018). In theory, the Type 1 gene set alone should be CI-capable (Lindsey et al., 2018); though Cif_wSuz[T1] mutations may be in key residues for CI expression. Notably, wRi does not cause CI when transinfected into D. suzukii (Cattel et al., 2018). Thus, the Cif_wSuz[T1] proteins may be able to cause CI, but their effects are inhibited by suppressors encoded in the D. suzukii genome. In a separate example, the triple-strain infection of wAlbA, wAlbB, and wMel in A. albopictus can cause CI but cannot self-rescue (Ant and Sinkins, 2018). Since each of the individual Wolbachia strains in this triple-strain infection can cause CI and are self-compatible in A. aegypti (Ant and Sinkins, 2018), neither genetic variation in the cif genes alone nor host suppressors can explain the emergence of self-incompatibility. However, it is plausible that some Wolbachia may inhibit the reproductive manipulations of other co-infecting strains, but this requires further testing. In summary, the relationships between Wolbachia strains and their hosts are likely to have an impact on CI. Additional work is necessary to answer these persistent questions: how does superinfection impact CI expression, how does the host act to suppress CI phenotypes, and what are the evolutionary dynamics that govern these interactions?

There is strong evidence for the genetic basis of unidirectional CI between Wolbachia-bearing and aposymbiotic arthropods (Figure 1A; Beckmann et al., 2017; Chen et al., 2019; LePage et al., 2017; Shropshire et al., 2018; Shropshire and Bordenstein, 2019). However, the genetic basis of bidirectional CI between arthropods with different symbiont strains remains poorly understood (Figure 1B,C). Sequence divergence in CI-associated factors has long been thought to be a contributing factor to these incompatibilities, namely that divergence in both CI-causing and rescue-causing genes would be required for bidirectional CI to emerge (Charlat et al., 2001). Indeed, phylogenetic analyses of cif genes reveal that strains carrying similar alleles tend to be compatible, strains with more distantly related cif genes are not, and a single Wolbachia strain can have multiple unique cif gene pairs (Bonneau et al., 2019; Bonneau et al., 2018a; LePage et al., 2017). For instance, when wMel is transinfected into a D. simulans background, it is unidirectionally incompatible with the native wRi strain, wherein wRi can rescue wMel-induced CI but the reciprocal cross is incompatible (Poinsot et al., 1998). Intriguingly, wRi carries Type 1 cif genes closely related to wMel’s and a divergent Type 2 gene pair. Thus, it is plausible that wRi can rescue wMel’s CI because of CifA_wRi[T1], whereas wMel cannot rescue wRi’s CI because it lacks a rescue gene for the Type 2 gene pair (LePage et al., 2017). Additionally, population genetic analyses of cif genes in wPip reveal that there are numerous unique strains, each strain carries multiple closely-related cif variants that belong to Type 1 and Type 4 cif clades, and a single genetic variant of CifB_wPip[T1] correlates with the inability of one strain of wPip to rescue CI caused by a divergent wPip strain (Atyame et al., 2011b; Bonneau et al., 2019; Bonneau et al., 2018a). However, while these data suggest that cif genetic variation and/or copy number contributes to strain incompatibility, it remains possible that the considerable host genotypic variation between these incompatible populations contributes to these relationships in a way that also correlates with cif genotypic diversity (Atyame et al., 2011a). More reductionist functional studies that control for variation in host genotype will be necessary to confirm that cif sequence variation alone can explain CI relationships.

Historically, CI and rescue were thought to be caused by different genes, and that divergence in both genes would be required for bidirectional CI to evolve relative to an ancestral strain (Charlat et al., 2001). Thus, this model for bidirectional CI requires two steps: one mutation for CI and one for rescue. A major limitation of this model is that the intermediate state, wherein only one of the two phenotypes have shifted, is self-incompatible and represents a ‘maladaptive valley’ unlikely to persist as a rare variant. Given the abundance of bidirectionally incompatible strains across the arthropod Wolbachia (Atyame et al., 2011b; Bordenstein and Werren, 2007; Branca et al., 2009; O'Neill and Karr, 1990; Sicard et al., 2014), and the rarity of so-called ‘suicidal’ self-incompatible strains (Zabalou et al., 2008), crossing this maladaptive valley may be an unlikely evolutionary scenario. In contrast, since CifA is involved in both CI and rescue, it becomes possible for a single mutation that affects CI to also impact rescue. Thus, a single mutation in CifA may shift both CI and rescue phenotypes, yield bidirectional CI relative to an ancestor, and maintain self-compatibility (Shropshire et al., 2018). Indeed, mutagenesis of highly conserved amino acids across CifA reveal that sites within CifA’s N-terminal region are crucially important for the expression of both CI and rescue, suggesting that residues in this single region are coopted for both phenotypes (Shropshire et al., 2020). Notably while this one-step model of bidirectional CI avoids the maladaptive valley, it may only spread if transferred into a new aposymbiotic (sub)population since emergence of a new incompatibility type within a symbiont-bearing population would be immediately incompatible with the more common symbiont in the population. More research is needed to fully understand the genetic basis of bidirectional CI and its evolution. For instance, theoretical modeling will be necessary to evaluate additional routes of emergence as it relates to Cif sequence variation, functional genetic assays can be used to unravel the correlation between cif sequence variation and (in)compatibility relationships, and population genetic surveys coupled with population dynamics modeling would reveal when a novel variant would be likely to persist in a population.

Structural homology-based analyses suggest that Type 1 CifA have three putative domains: a catalase-related (catalase-rel) domain involved in the degradation of reactive oxygen species, a domain of unknown function (DUF) 3243 with homology to a Puf-family RNA-binding domain (RBD), and a sterile-like transcriptional regulator (STE) (Figure 4A; Lindsey et al., 2018). While the catalase-rel domain is unique to the CifA_[T1], the STE is maintained in Type 1–4 genes (Lindsey et al., 2018), and the Puf-family RBD exists in Type 1–5 genes (Bing et al., 2020b; Martinez et al., 2020). Importantly, these annotations are of low predictive value (20–30% probability) and may not withstand experimental testing (Lindsey et al., 2018). On the other hand, sliding window analyses of selection for CifA_[T1] suggest that while the full protein is under purifying selection, the catalase-rel domain and the unannotated N-terminal region are under the strongest selection (Shropshire et al., 2018). Indeed, CifA cannot contribute to transgenic CI or rescue when conserved amino acids are mutated within the unannotated N-terminal region or in the putative catalase-rel domain of CifA_wMel[T1] (Shropshire et al., 2020). Conversely, when sites are mutated in CifA’s DUF domain, it maintains the ability to contribute to rescue, but loses CI capability (Shropshire et al., 2020). Thus, CifA’s N-terminal region is crucially important for both CI and rescue, whereas sites within the DUF domain are only crucial for CI (Figure 4A). More work will be necessary to determine how and why these mutations impact these phenotypes, but the annotations provided above afford a set of testable hypotheses and questions. For instance, does CifA interact with reactive oxygen species both in the context of CI and rescue, and/or does RNA-binding occur in the context of CI? Biochemical assays testing for these functions will further elucidate how CifA contributes to CI and rescue phenotypes.

Additionally, how CifA is involved in both CI and rescue remains largely a mystery. The simplest explanation is that CifA maintains the same function in both CI and rescue. Under this framework, CifA would act on a pathway that can be modified during spermatogenesis and oogenesis to produce opposite affects (Shropshire et al., 2018). If CifA were to drive such a function, then CifB’s role in CI would seemingly be auxiliary and perhaps only necessary for localization of CifA to particular targets or, since CifB acts as a deubiquitinase, to protect CifA from degradation by ubiquitin pathways (Beckmann et al., 2017). Alternatively, CifA may be a multi-functional protein that employs one set of functions to cause CI and another to cause rescue (Shropshire et al., 2018). For instance, if CifA targets sex-specific host pathways, CifA can only affect its host in a particular way if that target is available. Additionally, CifA may be modified in some manner that differs between the testes and the ovaries, unlocking unique biochemical functions by posttranslational modification, localization differences, or the expression of different protein conformational isoforms (Shropshire et al., 2018). In summary, little is known about CifA’s functional role in CI and rescue, and considerable work is necessary not only to identify its enzymatic capabilities, but also to further elucidate how it can act both to cause and prevent CI.

CifB_[T1] from both wMel and wPip encode a single putative ubiquitin-like protease (Ulp1) domain (Beckmann et al., 2017; LePage et al., 2017; Lindsey et al., 2018). The Ulp1 domain was later expressed separate from the rest of the protein in E. coli and purified for downstream in vitro ubiquitin cleavage assays (Beckmann et al., 2017). When exposed to a variety of ubiquitin chains, it was revealed that the Ulp1 cleaves K6-, K11-, K27-, K29-, K33-, K48-, and K63-linked ubiquitin in vitro, but with a preference for K63 chains (Figure 4B; Beckmann et al., 2017). K63 chains are associated with NF-κB signaling which has diverse functions including innate immunity, DNA transcription, autophagocytosis (Tan et al., 2008; Wertz and Dixit, 2010), and proliferation of cell nuclear antigen (PCNA) (Ripley et al., 2020) that has previously been shown to act abnormally in CI-affected embryos (Landmann et al., 2009). A single amino acid mutation in the catalytic site of the Ulp1 prevents the breakdown of ubiquitin chains in vitro (Beckmann et al., 2017). Expressing the Ulp1 catalytic mutant for CifB_wPip[T1] and CifB_wMel[T1] in male D. melanogaster alongside CifA did not induce CI, suggesting that deubiquitilase activity is important for CI barring the occurrence of any potential protein structural changes in the mutants (Figure 4A; Beckmann et al., 2017; Shropshire et al., 2020). However, some caution is warranted as deubiquitinase assays have not been conducted using the full-length protein, and it is unknown if this activity is maintained in vivo. Moreover, it remains unknown what CifB deubiquitinates, if anything, inside reproductive tissue cells and how this deubiquitination contributes to CI. It is also interesting that nuclear localization of the male-killing protein Spaid in the endosymbiont Spiroplasma poulsonii is impacted by a domain annotated as a deubiquitinase, suggesting that host ubiquitin regulation or localization to host nuclei is important for reproductive manipulation (Harumoto and Lemaitre, 2018). Future biochemical assays will help answer these persistent questions.

While CifB’s Ulp1 is seemingly important for CI, only CifB_[T1] have this domain. Moreover, additional mutagenesis assays reveal that other conserved sites across the CifB protein similarly ablate CI function (Figure 4A; Shropshire et al., 2020), strongly suggesting other regions of the protein are likewise important for CifB function. For instance, all CifB proteins (Type 1–5) are also annotated with a dimer of PD-(D/E)XK (hereafter PDDEXK) nuclease domains (Bing et al., 2020b; Lindsey et al., 2018; Martinez et al., 2020). Indeed, in vitro nuclease assays with CifB_wPip[T4] confirm that they can nick both double- and single-stranded DNA (Figure 4C; Chen et al., 2019). Moreover, mutating PDDEXK catalytic sites in CifB_wPip[T4] prevents nuclease activity in vitro and CI-inducibility when expressed in D. melanogaster (Chen et al., 2019). Unlike the other phylogenetic Types, CifB_[T1] proteins do not have the canonical PDDEXK catalytic sites, thus lending doubt to the importance of these domains as nucleases in CifB_[T1] (Beckmann et al., 2017). However, these domains remain structurally homologous to other PDDEXK domains (Lindsey et al., 2018), and many functional PDDEXK-like domains lack the canonical PD-(D/E)XK catalytic motifs, opting instead for alternative catalytic residues and structural folds. PDDEXK-like domains without catalytic sites are still involved in other DNA-associated processes (Knizewski et al., 2007). Mutating conserved amino acid residues in either of the PDDEXK domains of CifB_wMel[T1] inhibits its ability to contribute to CI (Shropshire et al., 2020). Additionally, despite wPip containing both Type 1 and 4 genes, there are no notable differences in cytological embryonic defects caused when both genes are expressed as compared to other strains that only have CifB_[T1], suggesting that these genes yield similar cytological outcomes (Bonneau et al., 2018b). Biochemical assays will be necessary to evaluate the nuclease activity of a diverse array of CifB proteins including CifB_[T1] because the conserved areas in and around the PDDEXK domains across all Cif Types likely persist because of a common function that underpins CifB’s involvement in CI.

A list of putative Cif protein-binding partners have recently been reported. While CifA_[TI] and CifB_[T1] bind in vitro (Figure 4D; Beckmann et al., 2017), it does, however, remain unknown if CifA binds CifB in vivo in the testes to cause CI or if maternal CifA binds to paternal CifB in the embryo to cause rescue (Beckmann et al., 2019a; Beckmann et al., 2019b; Shropshire et al., 2019). More work on the localization, co-localization, and binding profiles of these proteins will elucidate this question. Additionally, Cifs appear to bind to a suite of host proteins that differ based on if CifA and CifB are expressed alone or together. Sixty-seven host proteins were identified as Cif binding partners under co-expression of CifA and CifB, whereas 45 proteins were identified with CifB expression alone in pools of male and female D. melanogaster (Beckmann et al., 2019c). Karyopherin-α (Kap-α) is notable among these proteins. It bound to singly expressing CifB extracts, and its overexpression in aposymbiotic females yielded partial rescue (~20% hatch rate improvement) when crossed to transgenic CI males (Beckmann et al., 2019c). Kap-α is a nuclear import receptor and a regulator of p53 which has roles in the protamine-histone exchange process (Beckmann et al., 2019c; Emelyanov et al., 2014). Intriguingly, delayed H3 histone deposition is a hallmark of CI during early embryogenesis (Landmann et al., 2009), suggesting a relationship between CifB, Kap-α, p53 and histone-associated abnormalities in CI. However, it is important to emphasize that Kap-α was only pulled down when CifB was singly expressed (Beckmann et al., 2019c), suggesting that while Kap-α overexpression may influence rescue-efficiency, it is unclear how it would be part of the rescue mechanism since CifB is not necessary for rescue to occur. More work is essential to determine if CifB’s binding to Kap-α contributes to CI and how Cif binding to other host proteins relates to CI and rescue phenotypes.

Decades of research have characterized an in-depth understanding of CI-associated cytological abnormalities. These studies broadly define alterations during spermatogenesis (Figure 5A) and embryogenesis (Figure 5B), suggesting that CI is associated with a sperm modification prior to fertilization that results in embryonic defects and death. Importantly, the causes of the reported sperm and embryonic abnormalities remain unknown and, in most studies, it is unclear if these observations are directly related to the Cif proteins or are a byproduct of Wolbachia symbionts in the testes. However, these findings provide insight into the ways in which Wolbachia and CI influence host reproduction and fertility. Below, we review the cytological changes that occur in spermatogenesis and embryogenesis during CI and rescue, and we highlight areas where future research is crucially needed with reductionist assays to disentangle effects of Wolbachia symbiosis and CI.

Spermatogenesis is a highly regulated process. It begins with cells of the germline stem cell niche (GSCN) replicating into spermatogonia that subsequently undergo mitosis to yield a spermatocyst with 16 spermatocytes (Fuller, 1993; Hackstein, 1987; Lindsley, 1980). Each spermatocyte in the cyst then undergoes two rounds of meiosis to form four spermatids, for a total of 64 spermatids in each cyst. The spermatids then undergo elongation where the sperm tail forms, and histones are replaced with protamines for tight packaging of DNA in the nucleus within the sperm head (Rathke et al., 2014). In the final stages of sperm maturation, spermatids undergo individualization to remove excess cytoplasm before becoming mature sperm to enter the seminal vesicle for storage before mating. Impacts of Wolbachia on spermatogenesis can result in downstream sperm defects that may be connected to CI (Figure 5A). For example, symbiont-bearing D. simulans flies and Ephestia moths produced fewer sperm, and stronger CI was associated with more sperm transfer during copulation in D. simulans (Awrahman et al., 2014; Lewis et al., 2011; Snook et al., 2000). When D. simulans females mated with Wolbachia-bearing and aposymbiotic males, the sperm of aposymbiotic males were more likely to fertilize eggs (Champion de Crespigny and Wedell, 2006), suggesting that Wolbachia-modified sperm are less competitive. Wolbachia-affected sperm cysts exhibit abnormal morphology with some sperm fused together and other sperm exhibiting randomly-oriented, axoneme-mitochondrial complexes that are responsible for sperm motility (Riparbelli et al., 2007), perhaps explaining fertility defects and variation in sperm competition. However, key questions remain. Are cif gene products responsible for the aforementioned sperm abnormalities? If these defects are not caused by the cif genes, and are instead a byproduct of other Wolbachia-host associations, what is their significance, if any, to reproductive parasitism? In summary, a deeper investigation of these defects as they relate to CI products is needed to confirm their link to CI.

Interestingly, Wolbachia are not symmetrically distributed in testes, with only some spermatocysts harboring symbionts in the strong CI-inducing wRi strain of D. simulans (Clark et al., 2003). Indeed, wRi is almost exclusively localized to the GSCN, and some GSCN remain aposymbiotic, suggesting that the Cif proteins must either act early in spermatogenesis or are diffusible factors that can stably travel into later stages of spermatogenesis (Clark et al., 2003; Clark et al., 2002; Riparbelli et al., 2007). Wolbachia are stripped during the individuation process and moved into waste bags where they are presumably degraded (Riparbelli et al., 2007). Not only does this suggest that Wolbachia create a diffusible factor that interacts with sperm or spermatogonia to cause CI, but it also helps to explain why paternal Wolbachia transfer has not been observed (Yeap et al., 2016) with rare exceptions such as in hybrid Nasonia wasps and in transinfected A. aegypti (Chafee et al., 2011; Ross et al., 2020a). Future work investigating the localization of CifA and CifB will determine when and where the Cif proteins act to cause CI and whether they are transferred to the embryo for the potential to directly cause defects during embryogenesis.

Abnormalities that define the post-fertilization events underpinning CI are observed during embryogenesis (Figure 5B). In chronological order, CI-affected embryos experience abnormal maternal H3.3 histone deposition on the male pronucleus, delayed activation of the DNA polymerase cofactors PCNA and cell cycle regulator Cdk1 resulting in incomplete DNA replication, delayed nuclear envelope breakdown prior to the first mitosis, and a delay in the first mitotic event (Landmann et al., 2009; Tram and Sullivan, 2002). These defects often culminate in a chromatin bridging phenotype during the first mitosis, shredding the paternal nuclei and leading to embryonic arrest (Breeuwer and Werren, 1990; Callaini et al., 1996; Lassy and Karr, 1996; Reed and Werren, 1995; Ryan and Saul, 1968; Tram et al., 2006). Notably, it remains unknown what the most proximal event is during CI. It is plausible that abnormal histone deposition is the first CI-causing event during embryogenesis that leads to a cascade of effects culminating in the other embryonic abnormalities, but there remain open questions. For instance, how do the Cif proteins interact with the host to cause abnormal histone deposition? Are the Cif proteins even transferred with the sperm so that they can cause these defects directly, or are these defects caused by an initial Cif interaction occurring during spermatogenesis? Finally, if Cifs do directly cause abnormal histone deposition, how are these affects rescued by the presence of CifA expressed in the embryo? A combination of cytological, transgenic, and biochemical assays may be necessary to evaluate these questions.

Defects in the first mitotic division are traditionally viewed as a key cytological outcome of CI, but abnormalities later in embryogenesis are also common and increasingly appreciated (Bonneau et al., 2018b; Callaini et al., 1997; LePage et al., 2017; Ryan and Saul, 1968). There are three distinct phenotypes: early mitotic failures whereby embryonic arrest occurs after several successful rounds of division, regional mitotic failures where some regions of the embryo appear to be dividing without issue, and widespread chromatin bridging in later stages of division (LePage et al., 2017). The cause of these defects remains unknown, but at least two hypotheses can be proposed. First, late stage embryonic defects are caused by the same cascade of abnormalities that often cause arrest during the first mitosis. Under this scenario, the difference in the cytological outcomes of the embryo may be explained by the magnitude of the proximal CI-causing affect. For instance, strong male pronuclear delay can result in complete exclusion of the male pronucleus from early development, yielding an embryo that may attempt to undergo haploid development (Callaini et al., 1997; Tram et al., 2006). In N. vitripennis where haploid individuals become males and diploids become females, exclusion of the male pronucleus during CI can manifest in haploidization where even fertilized eggs develop as haploids (Bordenstein et al., 2003; Ryan and Saul, 1968; Tram et al., 2006; Vavre et al., 2001; Vavre et al., 2000). Thus, the intensity of pronuclear delay may correspond with the resulting phenotypic profile during embryogenesis, but more work is necessary to determine if these effects translate to late stage embryonic defects. Second, these later stage abnormalities may be independent from the defects preceding the first mitotic failure. Indeed, it has been proposed that different phylogenetic Types of Cif proteins may contribute to different cytological outcomes (Bonneau et al., 2018b). However, wPip, which encodes both Type 1 and 4 genes, display both early and late stage embryonic abnormalities comparable to wMel which encodes only Type 1 genes (Bonneau et al., 2018b; LePage et al., 2017). Despite presumably having different mechanistic bases, Cardinium and Wolbachia both have converged on similar outcomes during early embryogenesis, including chromatin bridging and abnormal number of chromosomes after the first division (Gebiola et al., 2017). It remains unknown if Cardinium CI yields comparable molecular defects and sperm abnormalities to Wolbachia-induced CI. Clearly, there are a diverse set of cytological outcomes associated with CI in both Wolbachia and Cardinium. Additional cytological and reductionistic studies will be necessary to evaluate the cause of this variation and determine how the Cif proteins contribute to these phenotypes.

Temperature is often correlated with CI strength and is likely to contribute to the dynamics that govern Wolbachia’s spread (Foo et al., 2019). High temperatures, usually exceeding 27°C, can have a significant negative impact on CI strength in Wolbachia-carrying A. aegypti (Ross et al., 2020b; Ross et al., 2019), T. urticae (van Opijnen and Breeuwer, 1999), D. simulans (Hoffmann et al., 1986), D. melanogaster (Reynolds and Hoffmann, 2002), A. scutellaris (Trpis et al., 1981; Wright and Wang, 1980), A. albopictus (Wiwatanaratanabutr and Kittayapong, 2009), and Nasonia (Bordenstein and Bordenstein, 2011). There is considerable evidence that high temperature impacts Wolbachia densities in various species including A. albopictus and A. aegypti (Foo et al., 2019; Ross et al., 2020b; Ross et al., 2019), N. vitripennis (Bordenstein and Bordenstein, 2011), and T. urticae (Lu et al., 2012). High temperatures even cure hosts of Wolbachia (Jia et al., 2009). In natural populations of the butterfly Zizeeria maha, Wolbachia densities vary with season, and climate change may be contributing to a decrease in symbiont frequencies in the tropics (Charlesworth et al., 2019; Sumi et al., 2017). Notably in N. vitripennis and T. urticae, decreased Wolbachia densities and CI strength have also been correlated with an increase in phage WO lytic activity with higher temperatures (Bordenstein and Bordenstein, 2011; Lu et al., 2012). Cooler temperatures at or below 19°C have also been associated with decreased CI in D. simulans and N. vitripennis (Bordenstein and Bordenstein, 2011; Reynolds and Hoffmann, 2002). As with warm temperatures, cooler temperatures also yield increased phage WO densities, decreased Wolbachia densities, and decreased CI strength in N. vitripennis (Bordenstein and Bordenstein, 2011). These data suggest a phage density model of CI wherein phage WO may respond to temperature extremes by increasing its replication and lysing bacterial cells, thus lowering overall Wolbachia densities and resultantly CI levels.

However, while robust support of this model is available in N. vitripennis (Bordenstein and Bordenstein, 2011), more work is necessary to test if it is generalizable to other Wolbachia strains. For instance, in contrast to the relationships described above, Wolbachia in some D. simulans lines (Clancy and Hoffmann, 1998) and Leptopilina heterotoma wasps (Mouton et al., 2006) replicate more quickly at warmer temperatures, and yet CI strength decreases. Thus, it is plausible that phage WO in these species have a different relationship with temperature than in N. vitripennis, and other yet undescribed factors inhibit CI. Moreover, in E. suzannae bearing Ca. hertigii, high temperatures also yield reduced Cardinium densities and lower CI strength (Doremus et al., 2019). However, Cardinium do not harbor a phage, and thus phage lysis cannot explain this relationship. Additionally, in this same system, cooler temperatures yield reduced Cardinium densities, but an increase in CI strength (Doremus et al., 2019). Thus, here, it seems that bacterial densities alone do not explain the cause of CI strength variation. It is plausible that the factors contributing to CI strength variation in Wolbachia and Cardinium differ, and comparative phenotypic studies will be necessary to evaluate the differences between these two systems. However, in systems where symbiont density correlates with CI strength, it is plausible that the proximal cause is a shift in CI gene expression that correlates with symbiont densities. Transcript and protein abundance assays of Wolbachia’s cif genes will help elucidate this relationship when accompanied with measurements of variable CI strength.

Other correlates of CI strength variation are related to male and paternal grandmother age (Awrahman et al., 2014; Layton et al., 2019; Reynolds and Hoffmann, 2002), male mating rate (Awrahman et al., 2014; de Crespigny et al., 2006), male developmental timing (Yamada et al., 2007), rearing density (Yamada et al., 2007), and nutrition (Clancy and Hoffmann, 1998). All of these factors are significantly impacted by the structure of the population, resource availability, or behavior. Below, we will systematically discuss what, if anything, is known about how each of these factors impact CI strength.

First, male age can be negatively correlated with CI strength. For example, wMel of D. melanogaster has nearly no impact on embryonic hatching when males are 3–5 days of age, but can induce significant CI when males are less than 2 days of age (Reynolds and Hoffmann, 2002). Similar results have been observed with Wolbachia in D. simulans and N. vitripennis, but to varying degrees (Breeuwer and Werren, 1993; Karr et al., 1998). Since Wolbachia densities decrease with male age in D. melanogaster, D. simulans, and N. vitripennis hosts (Binnington and Hoffmann, 1989; Breeuwer and Werren, 1993; Bressac and Rousset, 1993; Clark et al., 2002; Karr et al., 1998; Reynolds and Hoffmann, 2002; Riparbelli et al., 2007; Turelli and Hoffmann, 1995; Veneti et al., 2003; Weeks et al., 2007), it is perhaps unsurprising that age also correlates with CI. Moreover, of the factors associated with CI strength, age is also the only one that has been investigated in the context of cifA and cifB transcription, and does indeed decrease with age alongside Wolbachia densities (LePage et al., 2017).

Interestingly, while older males have fewer Wolbachia, older virgin females have more (Layton et al., 2019). In fact, when females are aged longer prior to mating, their male offspring are laid with higher Wolbachia densities and resultantly induce stronger CI (Layton et al., 2019). This phenomena has been termed the paternal grandmother age effect (PGAE) (Layton et al., 2019). It is unclear why age’s impact on Wolbachia density is sex-specific. However, the relationship between male age, symbiont densities, and CI strength may not be generalizable across all CI-inducing symbionts and their hosts. For instance, Cardinium of E. pergandiella cause CI that is unaffected by male age (Perlman et al., 2014), and studies disagree about the significance of the impact of age on CI caused by wRi of D. simulans (Awrahman et al., 2014; Binnington and Hoffmann, 1989; Bressac and Rousset, 1993). Thus, the impacts of age on symbiont densities and CI may be limited to some Wolbachia or alternatively, to particular host backgrounds. More comparative phenotypic work will be needed to understand the broader context of the relationship between age, CI strength, and cif expression.

Additionally, male mating rate is also negatively correlated with CI strength. For instance, symbiont-bearing D. simulans males mate more frequently than aposymbiotic males, and the increased mating rate yields weaker CI in later matings (Awrahman et al., 2014; de Crespigny et al., 2006). Symbiont-bearing males also transfer more sperm during copulation than aposymbiotic males during the first mating encounter, and decreased sperm transfer in subsequent matings corresponds with weaker CI (Awrahman et al., 2014). As such, the increased mating frequency may be a behavioral adaptation employed by some hosts to restore reproductive compatibility between symbiont-bearing males and aposymbiotic females (Awrahman et al., 2014). As with temperature and age (Bordenstein and Bordenstein, 2011; Layton et al., 2019; Reynolds and Hoffmann, 2002), it has likewise been hypothesized that Wolbachia densities may decrease upon remating (Awrahman et al., 2014), but this hypothesis has not been tested. Alternatively, it has also been hypothesized that the amount of time that sperm remains in contact with Wolbachia corresponds with how strong CI can be (Karr et al., 1998), thus remating may contribute to high sperm turnover that limits Wolbachia-sperm exposure. These hypotheses can be tested via Wolbachia density assays and microscopy of reproductive tissues upon remating.

Male development time is likewise correlated with CI strength. Here, wMel-bearing male D. melanogaster induce stronger CI when they are the first emerging males of a clutch (Yamada et al., 2007). The younger brothers, which take longer to develop but are approximately the same age, cause weaker CI. This phenotype has been termed the younger brother effect (YBE) (Yamada et al., 2007). The YBE is an outlier in phenotypes associated with CI strength variation in that younger and older brothers have comparable adult Wolbachia densities (Yamada et al., 2007), suggesting an alternative mechanism for the relationship between developmental timing and CI strength. However, it is also plausible that while younger and older brothers have similar bacterial densities, their localization may shift such that cells more important to CI expression have higher densities than other cells in the testes (Clark et al., 2002). Alternatively, Wolbachia densities of the adult male may be less informative than density differences during embryonic or larval development. For instance, the PGAE, as described above, revealed that sons of older females caused stronger CI and while their sons did not have higher Wolbachia densities as adults, they did have higher densities during embryogenesis (Layton et al., 2019). Intriguingly, Wolbachia densities rapidly declined in aged females after mating and embryo laying, suggesting that many Wolbachia were transferred from the ovaries to the developing egg and ultimately embryo. Thus, it is plausible that Wolbachia densities would correlate with deposition order such that first laid older brothers would have higher densities than younger brothers laid soon after (Layton et al., 2019). While these hypotheses remain associated and to be formally tested, it is also notable that the YBE does not appear to apply to wRi of D. simulans (Yamada et al., 2007), other studies have failed to replicate these results in other wMel-bearing D. melanogaster lines (LePage et al., 2017), and the opposite phenotype is observed with Cardinium of Encarsia where older brothers cause weaker CI (Perlman et al., 2014). Thus, additional work is necessary to replicate the YBE in D. melanogaster and other symbiont-host combinations and to evaluate its cause via longitudinal developmental studies of Wolbachia densities. Moreover, understanding why Cardinium and Wolbachia CI are differentially impacted by these factors is important in determining how symbiont dynamics relate to reproductive manipulation.

Finally, rearing density and nutrition can also impact CI strength relationships. For instance, when wMel-bearing D. melanogaster are reared in high densities, CI strength is lower than if they are reared in low densities (Yamada et al., 2007). The initial hypothesis behind this correlation was that high-density rearing led to nutritional stress which translated to less Wolbachia (Yamada et al., 2007). While this hypothesis has not been explicitly tested, there is a reasonable logical framework behind it. Indeed, multiple studies have shown that D. simulans males exposed to nutritional stress have weaker CI than males with abundant resources (Clancy and Hoffmann, 1998; Sinkins et al., 1995). Notably, nutritional stress is also correlated with reduced Wolbachia densities, supporting models, as above, where Wolbachia densities and cif expression are the proximal factors driving the relationship to CI strength. That said, it is important to note that recent microscopy studies have shown that standard qPCR-based measures of Wolbachia densities may not be adequate under nutritional stress since host ploidy is subject to variation based on diet (Christensen et al., 2019). Thus, it is plausible that qPCR-based variation in Wolbachia densities under nutritional stress may in fact be driven by variation in host ploidy and not Wolbachia density. Replication of these studies will be necessary to confidently link nutrition, rearing density, and Wolbachia densities to CI strength. Importantly, rearing density does not influence CI strength in wAlbA and wAlbB A. albopictus (Dutton and Sinkins, 2004), suggesting that even if wMel CI is impacted by rearing-density, this effect is perhaps not generalizable across Wolbachia-host combinations.

The factors described above do not work on CI in isolation but instead seem to be mingled in a state of perpetual complexity. For instance, the impact of temperature on CI strength in D. melanogaster is dependent on male age, where 1-day-old males reared at 25°C induce stronger CI than those reared at 19°C, but the inverse is true with 3- and 5-day-old males (Reynolds and Hoffmann, 2002). Moreover, age has a variable impact on CI strength in different host backgrounds, suggesting that genotypic variation in either the host or Wolbachia strain may impact these relationships (Duron et al., 2007a; Reynolds and Hoffmann, 2002). These studies highlight the complexity of Wolbachia-host-environment interactions and should motivate additional investigation to resolve the factors that underpin these variations and the host genetic loci that influence how impactful each factor might be in each host.

Relationships between Wolbachia phenotypes and host genotypes are frequently investigated through transinfections of a Wolbachia strain into a non-native background via injection (Hughes and Rasgon, 2014) or introgression of one species cytoplasm into another host’s background via repeated backcrossing (Chafee et al., 2011). For example, wMel Wolbachia of D. melanogaster traditionally cause weak CI (Holden et al., 1993), but induce consistently strong CI when transinfected into either D. simulans or A. aegypti (Poinsot et al., 1998; Walker et al., 2011). Similar results were also observed when wTei which induce weak or no CI in the D. yakuba complex (Charlat et al., 2004; Cooper et al., 2017; Martinez et al., 2020; Zabalou et al., 2004), are transferred into D. simulans (Zabalou et al., 2008). Moreover, despite seemingly carrying the same Wolbachia, different genetic lineages of the wasp N. longicornis express different compatibility relationships with other strains (Raychoudhury and Werren, 2012), and wVitA of N. vitripennis causes weak CI in its native host but strong CI when introgressed into N. giraulti (Chafee et al., 2011). Intriguingly, this affect is only observed with wVitA and does not apply to wVitB, which also causes CI, suggesting that both host and Wolbachia genetics play a role in CI phenotypes. These studies support models that predict hosts will be selected to develop resistance against CI (Prout, 1994; Turelli, 1994), and raise many questions about an evolutionary arms race between Wolbachia and its host to control reproductive parasitism.

There are at least two broad models for the mechanisms of host suppression of CI: divergence of host products that are targets for CI (defensive model) or evolution of host products that inhibit and suppress Wolbachia or CI products (offensive model). First, a defensive model would predict that the pathway(s) that CI act(s) on in the host must be conserved enough for CI to be transferable between species, but also malleable enough for the pathway(s) to become resistant to CI. The host genes, transcripts, and proteins described earlier in this review are excellent candidates since they can mimic CI phenotypes (Biwot et al., 2020; Huang et al., 2019; Ju et al., 2017; Yuan et al., 2015; Zheng et al., 2011). However, studies are necessary to investigate genetic variation in these host products to assess the possibility that they are under selection to suppress CI. Other candidates would be direct binding partners of CifA and CifB (Beckmann et al., 2019c), but nothing is known about how natural variation in these products may relate to CI suppression. Conversely, an offensive model could yield the evolution of host genes involved in Wolbachia density regulation or some other target with indirect effects on CI strength. Notably, since these products may not be involved in the CI mechanism, they would not necessarily be expected to be conserved. For instance, the Wolbachia density suppressor (Wds) gene of Nasonia acts to suppress densities of wVitA, is taxon restricted to bees and wasps, and is under positive selection as would be expected for a suppressor acting in an evolutionary arms race with Wolbachia (Funkhouser-Jones et al., 2018). Since Wds is only present in Hymenoptera, it clearly is not generalizable as a standard mechanism of Wolbachia and CI suppression. However, it is plausible that other hosts have converged on comparable mechanisms of CI suppression. Additional research will be needed to reveal the diversity of mechanisms surrounding CI suppression and to understand the dynamics controlling their evolution.

The mod/resc model defines a mod factor as a CI-inducing product produced in males and a resc factor as a rescue-inducing product produced in females (Werren, 1997). The mod/resc model is agnostic to the genetic, biochemical, enzymatic, or cytological basis of CI. Instead, the mod/resc model provides a framework for describing the phenotypic expression of different Wolbachia strains. For example, a standard CI-inducing strain that can self-rescue would be denoted as mod+/resc+. Less common phenotypes include so-called suicidal Wolbachia (mod+/resc-) and Wolbachia that do not cause CI but can rescue CI induced by other strains (mod-/resc+) (Ant and Sinkins, 2018; Meany et al., 2019; Zabalou et al., 2008). Wolbachia that do not cause CI or rescue are designated mod-/resc-.

The mod/resc model assumes that for bidirectional CI to occur, the mod and resc factors would differ in such a way that they remain functional but are incompatible with each other (Charlat et al., 2001; Werren, 1997). As such, a strain can carry multiple mod or resc factors that determine the compatibility relationships with other strains, and the mod/resc model can be used to estimate the number of mod and resc factors within a host (Zabalou et al., 2008). To do this, Wolbachia strains are transinfected or introgressed into the same genetic background and then crossed to determine the incompatibility relationships between strains or against aposymbiotic flies. A strain that causes CI against an aposymbiotic female is considered to have at least one mod factor. If it can rescue itself then it has at least one resc factor. If two CI-inducing and self-compatible strains are bidirectionally incompatible, then it is assumed that each carry at least one set of mod and resc factors but that they are not the same. Indeed, crossing experiments between various Wolbachia strains have revealed unidirectional and bidirectional incompatibilities which have led to agreement that Wolbachia frequently carry multiple mod and resc factors (Poinsot et al., 1998; Zabalou et al., 2008).

With the identification of the CI and rescue genes (Beckmann et al., 2017; Chen et al., 2019; LePage et al., 2017; Shropshire et al., 2018; Shropshire and Bordenstein, 2019), it is compelling to abandon the mod/resc model in favor of a purely genetic description of CI relationships. With the ever-growing availability of genomic datasets, acceptance of a gene-centric analysis of CI may be the simplest way to predict CI of a symbiont. However, while sequence information can indeed yield informed hypotheses about a strain’s CI, some hosts suppress their symbiont’s CI (Chafee et al., 2011; Poinsot et al., 1998; Walker et al., 2011), and some symbiont strains exhibit different forms of reproductive parasitism based on their host background (Fujii et al., 2001; Jaenike, 2007; Sakamoto et al., 2005; Sasaki et al., 2002; Zabalou et al., 2008). Thus, we propose that a modern framework for describing CI relationships should involve both phenotypic data described under the mod/resc model and genetic data described under the Two-by-One model. For example, if genomic sequencing of a novel Wolbachia precedes phenotypic observations, then a genetic analysis could reveal cifA and cifB homologs that are either comparable to those in CI-inducing strains or contain putative loss-of-function mutations. Phenotypic data is of course necessary to confirm the hypothesis. Indeed, wYak of D. yakuba and wRec of D. recens cause CI, but they have cifB genes with stop codons that truncate the proteins relative to the wMel and wPip cifB (Cooper et al., 2017; Martinez et al., 2020; Shoemaker et al., 1999). As such, a genetic description of these strains alone could result in mischaracterization of wYak and wRec as non-parasitic strains with putative cifB pseudogenes. It is only with knowledge of both cif gene sequence and phenotypic data that a complete understanding of the basis of CI in these strains is possible. Thus, both the Two-by-One and mod/resc models will serve as a useful framework to describe these systems.

HM models (Figure 7A) make at least two key predictions. First, male host products are modified by Cifs (Shropshire et al., 2019). There are numerous, pre-fertilization defects associated with CI-inducing Wolbachia including changes in sperm morphology and competitive ability (Champion de Crespigny and Wedell, 2006; Riparbelli et al., 2007), supporting that the host is modified prior to fertilization. It is unknown, however, whether these outcomes are due to pre-fertilization defects causally related to CI or general responses to Wolbachia in the testes. Second, and most crucially, the proximal CI modifications causing death of the fertilized embryo is rescued (e.g., replaced or otherwise negated) by CifA in the embryo (Shropshire et al., 2018; Shropshire and Bordenstein, 2019). CifA does not rescue through binding with male-transferred CifB products since Cif protein yields the modification prior to fertilization in testes and thus does not need to be transfered with the sperm to the embryo. Instead, CifA may interact with host processes to reverse or otherwise stop the effects of CI caused by CifA and CifB protein expression in males. As such, assessment of the location of CifA and CifB binding (testes or embryo), the transfer of Cif products, if any, with the sperm, and the interactions that Cif have with the host will further inform this model. We discuss three additional non-exclusive HM-based models below: titration-restitution, mistiming, and goalkeeper.

The titration-restitution model (a.k.a. the ‘sink’ hypothesis) was originally proposed by Werren, 1997 and posits that CI is induced by over- or under-expression of host products or pathways in the testes/sperm and rescue occurs when the same products are misregulated in the opposite direction in the ovaries/embryo (Figure 8A; Kose and Karr, 1995; Poinsot et al., 2003; Werren, 1997). Indeed, Wolbachia have a considerable impact on expression profiles, some host genes are differentially expressed in male and female reproductive tissues, and numerous host factors meet these criteria (Baldridge et al., 2017; Baldridge et al., 2014; Bing et al., 2020a; Yuan et al., 2015), as described in the section above. There are at least two ways in which CifA and CifB proteins can underpin the titration-restitution model. First, since CI and rescue would occur through titration of the same host product or pathway, it is feasible that CifA, which acts on both sides of the phenotype (Shropshire and Bordenstein, 2019), may drive these expression changes. Under such a model, CifB may act as an ‘accessory protein’ that enables CifA to target a paternally derived product that it would otherwise not be able to reach on its own. Second, CifA may act on its own to up- or down-regulate host products but has the opposite impact on that product when CifB is present. As such, rescue would occur through CifA’s lone action which counteracts the misregulation caused by CifA and CifB dual expression.

Notably, titration-restitution models can explain bidirectional CI if Cif products from different strains have variable impacts on multiple host expression pathways. Thus, rescue would not be possible from a second strain since it could be targeting the wrong host factor or pathway. Indeed, divergent CI genes may differentially impact host pathways. For example, only the CifB_[T1] sequences maintain a functional Ulp1 domain while the other four CifB clades have a dimer of PDDEXK nucleases that is also present in CifB_[T1] (Beckmann et al., 2017; Bing et al., 2020b; Lindsey et al., 2018; Martinez et al., 2020). It is feasible that CifB with different domains impact different host pathways. Alternatively, Cif proteins may have differential impacts on the level of misregulation instead of or in addition to impacting multiple host pathways which may influence incompatibility relationships. More work will be necessary to understand if cif expression influences transcriptional and translational variation and how that variation corresponds to CI.

The mistiming model (a.k.a. the ‘slow motion’ hypothesis) was first explicitly proposed by Tram and Sullivan, 2002 and is based on the observation that the paternal pronucleus has slowed development relative to the female pronucleus in CI crosses, the rescue cross has normal cell cycle timing, and the female pronucleus continues development despite the slowdown in both Drosophila and Nasonia species (Figure 8B; Callaini et al., 1996; Ferree and Sullivan, 2006; Ryan and Saul, 1968; Tram and Sullivan, 2002). This established the hypotheses that delayed male pronuclear development is responsible for emergent defects in early embryogenesis, and that resynchronization of the development may occur by comparably slowing down the development of the female pronucleus or slowing the cell cycle in rescue. Since the cell cycle timing of the female pronucleus is what establishes the timing for the first mitosis (Bossan et al., 2011), symbiont-bearing females do not induce CI because the male pronucleus reach apposition prior to the female pronucleus. Though, the reciprocal cross would be incompatible because the female pronucleus finished development prior to the male, and the first mitosis would have initiated before the male pronucleus arrives. Importantly, this model predicts that CI crosses are subject to haploidization of diploid offspring since the male pronucleus could be completely excluded from mitosis if it was significantly slowed. This is indeed the case in N. vitripennis where CI often manifests as only male offspring since haploid offspring are viable in this species but develop as males (Bordenstein et al., 2003).

The mistiming model proposes that CI and rescue have comparable impacts on the development of male and female gametes, respectively. As such, a single gene could in theory be responsible for both CI and rescue (Poinsot et al., 2003). Under this paradigm, CifA may enact a slowdown in both tissues since it is involved in both phenotypes (Shropshire and Bordenstein, 2019). However, if this were the case, then what would be the purpose of CifB? It is possible that CifB is responsible for localizing CifA to a male-specific target where it imposes the same outcomes on its host. Since this hypothetical male product would not be available in the embryo, CifB would not have a role in rescue. However, an alternative model for mistiming is that rescue may not occur through slowing down the female pronucleus but may instead work by removing the slowdown agents from the male pronucleus. Together, these models would help to explain the proximal cause (misregulation) and culminating effects (mistiming) of CI. More work will be necessary to understand if rescue occurs via slowdown of the female pronucleus or from speeding-up the male pronucleus.

A major limitation of the mistiming model is that it cannot explain bidirectional CI. Since mistiming proposes that rescue happens through delaying the female pronucleus as much as or greater than the male pronucleus, a sufficiently strong delay should yield compatibility with any strain that has a weaker male delay. As such, only unidirectional CI should manifest between strains where the strain inducing the stronger delay is capable of rescue. The goalkeeper model was proposed in 2011 as a way to address this limitation (Figure 8C; Bossan et al., 2011). In addition to the expectations of the mistiming model, goalkeeper suggests that a secondary factor unassociated with this mistiming may also be involved in CI. The combined contributions of these two mod factors leads to CI. Under this paradigm, CifA and CifB may contribute to different kinds of defects during spermatogenesis, each contributing in somewhat independent ways to CI. Rescue must then negate the impacts of both factors. Thus for CifA to rescue CI, it would not only need to contribute to a delay in the pronuclear development but also reverse the impacts of a secondary source of modification. Notably, since the titration-restitution model does not make predictions about the developmental timing of the male and female pronuclei, it is compatible with both mistiming and goalkeeper models and can help explain mistiming through misregulation of host factors in a manner that leads to slowed development. More functional genetic, biochemical, and cytological studies are necessary to understand how a goalkeeper model and/or a combination of these HM-based models may contribute to CI.

Since Wolbachia are not paternally inherited, Hurst proposed in 1991 that Wolbachia make a CI-inducing toxin that diffuses into the sperm cytoplasm and is transferred to the egg during fertilization and causes death (Hurst, 1991). Rescue then occurs when Wolbachia in the egg produce an antidote that binds to the toxin and prevents it from killing the embryo (Hurst, 1991). This TA model (Figure 7B) makes two key predictions (Beckmann et al., 2019a; Shropshire et al., 2019). First, the Cif proteins are transferred to the embryo. Mass spectrometry of spermatheca from symbiont-bearing Cu. pipiens females mated with symbiont-bearing males revealed fragments of CifA (Beckmann and Fallon, 2013). These later data have been used to support this prediction, but since these females harbor Wolbachia (Beckmann and Fallon, 2013) and CifA is also the rescue protein (Chen et al., 2019; Shropshire et al., 2018), the most parsimonious explanation for CifA’s presence in symbiont-bearing spermatheca is related to Wolbachia in females and potentially rescue, not CI. It remains possible that Cif proteins are transferred, but this is not the simplest interpretation of currently available data. Second, if the proteins are transferred, then maternal CifA must bind to the CI toxin to prevent function. In vitro biochemical assays reveal that CifA and CifB are capable of binding (Beckmann et al., 2017), but it remains unknown if they bind as a toxin complex to induce CI or if CifA binds to CifB in the embryo to rescue CI. Moreover, while CifB’s Ulp1 domain is an in vitro deubiquitinase, CifA’s binding to CifB does not inhibit deubiquitinase activity, suggesting that if binding is for the purpose of rescue it is not inhibiting one of CifB’s biochemical functions (Beckmann et al., 2017). As such, assays investigating if the Cif products are transferred to the embryo at all and where the Cif proteins bind each other in reproductive tissue cells will inform the foundation of this hypothesis.

The TA model traditionally states that the toxin and antidote are separate factors (Poinsot et al., 2003). However, our genetic understanding is that CifA is involved in both CI and rescue. There are two ways to update the model to be consistent with a Two-by-One genetic framework (Shropshire and Bordenstein, 2019) while maintaining the key assumptions of the TA model (Hurst, 1991). First, CifB may be the sole toxin but requires CifA as an antidote even during spermatogenesis to prevent overly defective sperm (Beckmann et al., 2019a). For this to work, CifA is expected to degrade faster than CifB, leaving CifB alone to enter the egg as a toxin unless it binds to maternally-derived CifA (Beckmann et al., 2019a). Alternatively, CifA and CifB could work together as a toxin complex that enters the embryo and is then rescued by maternallyA. Binding assays coupled with microscopy and localization studies will reveal when and where CifA acts relative to CifB.

As described above, the TA model aims to explain unidirectional CI between symbiont-bearing and aposymbiotic individuals. A modification of the TA model, called lock-and-key, expands the TA model to explain incompatibilities between Wolbachia strains. The lock-and-key model, like TA, proposes that a toxin is transferred from symbiont-bearing males to the embryo and will cause embryonic death unless an antidote is supplied. Toxins in this case are called locks, and antidotes are keys. The toxin lock is proposed to bind to or otherwise interfere with factors associated with proper embryonic development unless the antidote key is available to remove the lock. Bidirectional CI can then be explained by one strain carrying a set of locks and keys that are not compatible with the other strains’ locks and keys because of differences in binding affinity. This model leveraged predictions of the mod/resc model that strains can have multiple sets of mod/lock and resc/key factors and that a key is more likely to bind to its associated lock than to a divergent lock. Indeed, Wolbachia exhibit considerable cif polymorphism (Bing et al., 2020b; Bonneau et al., 2019; LePage et al., 2017; Lindsey et al., 2018; Martinez et al., 2020) and binding of CifA and CifB is strongest between cognate partners (Beckmann et al., 2017). However, the lingering questions with the TA model also apply with the lock-and-key model. Additionally, validation that divergent Cif proteins are functional, that they have differential impacts on the host, and contribute summatively to incompatibilities are lacking.