Mechanisms of action of Coxiella burnetii effectors inferred from host-pathogen protein interactions

Coxiella burnetii is an obligate Gram-negative intracellular pathogen and the etiological agent of Q fever. Successful infection requires a functional Type IV secretion system, which translocates more than 100 effector proteins into the host cytosol to establish the infection, restructure the intracellular host environment, and create a parasitophorous vacuole where the replicating bacteria reside. We used yeast two-hybrid (Y2H) screening of 33 selected C. burnetii effectors against whole genome human and murine proteome libraries to generate a map of potential host-pathogen protein-protein interactions (PPIs). We detected 273 unique interactions between 20 pathogen and 247 human proteins, and 157 between 17 pathogen and 137 murine proteins. We used orthology to combine the data and create a single host-pathogen interaction network containing 415 unique interactions between 25 C. burnetii and 363 human proteins. We further performed complementary pairwise Y2H testing of 43 out of 91 C. burnetii-human interactions involving five pathogen proteins. We used the combined data to 1) perform enrichment analyses of target host cellular processes and pathways, 2) examine effectors with known infection phenotypes, and 3) infer potential mechanisms of action for four effectors with uncharacterized functions. The host-pathogen interaction profiles supported known Coxiella phenotypes, such as adapting cell morphology through cytoskeletal re-arrangements, protein processing and trafficking, organelle generation, cholesterol processing, innate immune modulation, and interactions with the ubiquitin and proteasome pathways. The generated dataset of PPIs—the largest collection of unbiased Coxiella host-pathogen interactions to date—represents a rich source of information with respect to secreted pathogen effector proteins and their interactions with human host proteins.

for each effector and combined the data from both screens. However, we discarded single hits, i.e., interactions based on only one positive yeast colony.
All bait proteins were mapped to their corresponding C. burnetii locus tags, and all prey proteins were mapped to their official gene symbols as defined in the HUGO Gene Nomenclature Committee database [35] or Mouse Genome Informatics database [36]. When identified prey proteins could not be mapped to protein-coding sequences, we removed the interactions involving them. Moreover, we removed protein interactions between C. burnetii and "sticky" host proteins known to be indiscriminate binders as listed in S1 Table. Complementary pairwise Y2H testing of human-C. burnetii highthroughput protein interactions To test C. burnetii-human protein interactions identified by the Y2H library screening, we selected 94 pairs (involving 82 human proteins) for pairwise Y2H assay testing. We randomly selected these effector-host interactions among four relatively uncharacterized C. burnetii effector proteins (CBU0794, CBU0881, CBU1724, and CBU2078) and the plasmid protein CBUA0014 based on their large number of observed interactions in the high-throughput screens. We constructed the human prey Y2H clones by sub-cloning the ORFs from the Human ORFeome collection [37] into pGADT7g and pGADCg Y2H prey vectors. We successfully cloned 79 of the 82 human ORFs into Y2H prey vectors. We transferred the Y2H prey clones into yeast strain Y187 (MAT-a), and tested 81 of the 94 selected protein interactions, using the same procedure as outlined above to identify two-hybrid positive interactions in this screen.

Creation of expanded human-C. burnetii protein interaction network
We extracted human-murine orthologs from the NCBI HomoloGene database of homologs (www.ncbi.nlm.nih.gov/homologene) [38] and used them to identify orthology-based human-C. burnetii protein interactions [24]. We added the predicted orthology-based protein interactions to the human-C. burnetii Y2H data to create an expanded set of C. burnetii-human protein interactions. We used both human and murine libraries to provide better coverage of interactions with pooling of data allowing us to do a more robust statistical analysis. The data at hand do not support a statistical analysis of either species alone as meaningful. The provided data in the Supplementary Materials provide the species-distinct human and murine interactions.

Gene Ontology and pathway enrichment analysis of host genes
We performed standard enrichment analyses for C. burnetii-interacting host proteins as described previously [26]. Briefly, the enrichment of Gene Ontology (GO) [39] and Kyoto Encyclopedia of Genes and Genomes (KEGG) [40] pathways was calculated in R by using the Bioconductor packages BioMart [41] and KEGGgraph [42]. The background set of proteins for the GO analysis involved all constituent proteins from the human PPI network, and we used the complete GO tree annotation, excluding the root and the top two levels of GO terms. The background set of proteins for the KEGG enrichment analysis involved human proteins available in KEGGgraph that participated in at least one KEGG pathway. We used the Benjamini-Hochberg method [43] to correct all obtained p-values (p raw ) to adjusted p-values (p adj ).

High-throughput Y2H screening of host-pathogen interactions
We successfully cloned and prepared 33 C. burnetii effectors, and tested them in Y2H assays against both human and murine whole proteome libraries. Table 1 summarizes the available effector information, number of interacting host-pathogen proteins of each effector protein with either human or murine proteins, and protein interactions common to the two libraries. Of the tested pathogen proteins, 25 (76%) tested positive for interactions with either human or murine proteins, 12 (36%) showed interactions with both hosts, and eight failed to show any positive hits in our screens. The protein interaction data consisted of 273 unique interactions between 20 C. burnetii and 247 human proteins, and 157 between 17 C. burnetii and 137 murine proteins. The majority of C. burnetii proteins interacted with unique host proteins, i.e., 228 (92%) human proteins and 123 (90%) murine proteins interacted with a single C. burnetii protein. C. burnetii proteins that interacted with multiple proteins from either host also tended to interact with the other host. Of the nine pathogen effectors with more than 10 interacting host proteins, seven interacted with both hosts (7/9 or 78%). Of the remaining 16 that interacted with 10 or fewer host proteins, only five interacted with both hosts (5/16 or 31%). We observed far fewer instances of individual host-pathogen PPIs in both libraries; the Y2H screens identified five conserved PPIs between the human and murine data sets, i.e., interactions in which human proteins interacted with the same C. burnetii proteins as their murine orthologs. These differences could be due to low quantities of a prey gene in one of the two cDNA libraries or to non-exhaustive sampling of host-prey and pathogen-bait protein interactions.
Although Weber et al. reported that CBU0041 and CBU0885 were toxic to yeast when their expression was strongly induced in yeast [16], these proteins were not toxic in our Y2H screening using Saccharomyces cerevisiae AH109 and Y187 strains. The yeast Y2H vectors we used have a "2-micron" (2μ) origin where an endogenous (ADH1) promoter-resulting in a lowlevel expression of the recombinant protein-drives the protein expression and, hence, resulting in a lack of toxicity in our experiments.
The Y2H screens failed to capture the previously identified interaction between AnkG (CBU0881) and p32 [17]. A Y2H failure to detect (false negative) may have multiple origins, such as lack of exhaustive screening, true absence under the current set of experimental conditions, or deficiency of the target protein in the library (p32 was not detected in any other interaction in our screens). Currently, we are not able to determine the definitive cause of this absence.
In the following sections, we used the generated data to broadly characterize possible hostpathogen interaction phenotypes by performing overall analyses that take into account sets of observed and pooled interactions. These analyses do not allow us to individually account for each protein interaction, because individual protein-protein interactions range from strong binding events to more ephemeral signaling events and, as such, are not fully characterized or distinguishable by the deployed Y2H technique.

Orthology-derived high-throughput human-C. burnetii protein interaction network
The set of 25 interacting proteins represents a large fraction of potentially important effectors with a role in establishing the Coxiella infection. The total set of all interactions represents an interaction profile of multiple pathogen-targeted processes used to establish and maintain the bacterial infection. To characterize this interaction profile, we merged the human-C. burnetii experimental and orthologous data sets to create an expanded set of human-C. burnetii pairwise PPIs consisting of 415 unique interactions between 25 C. burnetii and 363 human proteins. Fig 1 graphically shows the resulting host-pathogen protein interaction network, with all interactions provided in S2 Table. Protein domain analysis of targeted host proteins We investigated the presence of conserved protein domains in host genes for nine C. burnetii effectors that interacted with 10 or more host proteins. Overall, 19 statistically over-represented conserved domains were identified among the effector-interacting host proteins. Table 2 shows that each effector has at least one statistically significant domain over-represented in the protein set compared to all host proteins, with over-represented host-domains typically present in 15% or less of the targeted host proteins for each effector. This indicated The binary interactions for all human and orthologous murine proteins are detailed in S2 Table. https://doi.org/10.1371/journal.pone.0188071.t001 that no single domain dominated the effector-host protein binding. The major exception to this was the calcium-binding EGF-like domain (EGF_CA), which appeared in 26 host proteins targeted by CBUA0014 and in 12 targeted by CBU0781. This domain is present in a large number of membrane-bound proteins as well as extracellular proteins and is essential for numerous protein-protein interactions [44]. Furthermore, CBUA0014 also targeted two host domainsthe PDZ_signaling domain among 10 host proteins, and the TGF-beta binding (TB) domain among nine host proteins. These domains are commonly identified as being part of multiple host protein-protein interaction events in organizing signaling complexes at cellular membranes that regulate cell proliferation, differentiation, and growth.

Enrichment analysis of high-throughput Y2H host-pathogen interactions
The relatively large number of known effectors studied and host-protein interactions retrieved allowed us to analyze the combined interaction data as an "effector profile" of coordinated interactions to identify a potential repertoire of host functions targeted by C. burnetii. This approach was intended to generate broad hypotheses on common underlying effector mechanisms. Hence, we used all available high-throughput Y2H data to identify pathways, biological processes, and cellular locations associated with the targeted host proteins. We used these data to construct a single host-pathogen protein interaction network, based on murine/human orthology, containing 415 unique interactions between 25 C. burnetii and 363 human proteins. Green nodes represent C. burnetii proteins, whereas pink and red nodes represent host proteins. Twelve C. burnetii proteins interacted with both hosts, three of which participated in conserved interactions (i.e., they interacted with both human proteins and their murine orthologs; shown as red nodes and connected with thick grey edges). https://doi.org/10.1371/journal.pone.0188071.g001

Int_alpha
Integrin alpha (beta-propeller repeats); found in adhesion molecules that mediate cellextracellular matrix and cell-cell interactions We performed domain identification, using NCBI CD Search (https://www.ncbi.nlm.nih.gov/Structure/bwrpsb/bwrpsb.cgi?) with default parameter settings for nine Coxiella burnetii (CB) genes that interacted with at least 10 host genes. 2 N d : number of occurrences of the domain among the targeted proteins; only domains that occurred at least three times in the host genes were analyzed. 3 N h : number of host genes that have the domain. 4 N db : number of occurrences of the domain in the background set. All representative human proteins that have been manually reviewed were downloaded from UniProt (http://www.uniprot.org/). Among the 20,201 genes, we used the 19,036 genes containing domains (NCBI CD Search) as the background set for the statistical analyses. 5 Original p-value from Fisher's exact test. 6 Adjusted p-value using the Benjamini-Hochberg correction procedure.  Table 3 shows the enriched KEGG pathways targeted by the effectors, using the orthologyderived high-throughput human-C. burnetii protein interaction network. This analysis indicates that the bacterial effectors have a preferential association with protein processing in the endoplasmic reticulum (ER), focal adhesion, and interference with protein degradation via the ubiquitin-proteasome pathway. The targeted host proteins affect glycolipid metabolism by interacting with host metabolism pathways involving small-branched hydrophobic amino acids. Finally, the effectors targeted host proteins involved in signaling via kinase regulation in the TGF-β and PI3K-AKT pathways. Table 4 shows the enriched GO Biological Process terms associated with the host proteins identified by the Y2H screen of the 25 interacting bacterial proteins. These terms are largely compatible with the KEGG analysis in Table 3, and highlight specific processes associated with metabolism and ubiquitin processing. The GO analysis also highlights immune response modulation processes involving immune receptor signaling and antigen presentation processes. Additionally targeted host functionalities include posttranscriptional regulation of gene expression. Table 5 shows the cellular localization of the host proteins. Overall, they were located in multiple cellular compartments in the cytoplasm, membrane-bound organelles, and nucleus. We could not identify any specific sub-nuclear location or particular intracellular vesicle for the host-targeted proteins, suggesting that they are present at multiple sites and potentially involved in multiple processes. The large number of host targets associated with extracellular exosomes points to processes associated with vesicles in general, as well as those with the potential to interact with the content of the exosomes as a means to influence the host immune response [45]. The locations associated with focal adhesions, adherence junctions, and microtubule cytoskeletons point to a preference for influencing cell signaling and vacuolar rearrangements associated with the infection. The association of targeted proteins with the ribonucleoprotein complex is consistent with a potential role for effectors in interfering with host protein processing in the ER. Additionally, a number of targeted proteins were located in the mitochondrial matrix.

Emerging interaction patterns from patterns of host interactions
Coxiella has the capability to successfully infect different eukaryotic hosts and cell types by using sets of translocated T4SS effector proteins. This implies that the interactions may be non-specific yet concerted to establish the biological host phenotype amenable to pathogen survival. The mechanisms by which Coxiella manipulates host cell processes are largely unknown, but effector-targeted host proteins can provide a mechanistic understanding of pathogenesis.
Our pathway analysis identified a number of metabolic host target proteins involved in small hydrophobic amino acid degradation and glycerolipid metabolism ( Table 3). Given that metabolism has not been considered as a host target process of T4SS effectors before [46], this suggests novel mechanisms involving energy metabolism (triglycerides are primarily used for energy storage) and essential amino acids (valine, leucine, and isoleucine), which may be related to preventing host cell autophagy by blocking a host-defensive starvation response [47].
The ability of C. burnetii to orchestrate physiological processes of host-cell organelles and interfere with host protein processing was evident from the large number of protein binding Mechanisms of action of Coxiella burnetii effectors inferred from host-pathogen protein interactions events that preferentially could take place in the ER and Golgi (Tables 3-5). Fig 2 shows the intervention points of nine screened Coxiella effectors affecting human-host protein processing in the ER. Previous studies have noted the importance and occurrence of pathogen interactions with the ER as a critical component of lipid metabolism, protein synthesis, protein trafficking, and cellular stress responses [48,49]. Fig 2 illustrates how both a single effector (CBU0794) interacted with multiple host proteins as well as multiple pathogen proteins targeting individual host proteins (Hsp40 and ERManI). The coordinated interactions affected processes, such as protein export, COPII-mediated vesicle formation, initiation of apoptosis, and ER-assisted degradation, via the ubiquitin-proteasome system. The recently noted interactions between the CCV and the ER, which involve the host protein oxysterol-binding protein homologue ORP1L and RAS oncogene RAB7A [49], were partly captured in our data with CBU0794 interacting with the oxysterol binding protein OSBPL1A. The summaries of host protein interactions in Tables 3-5 implicated multiple coupled host pathways involving focal adhesion, immune signaling and response (primarily via ubiquitinproteasome degradation), and changes in cell morphology. The main host targets in the focal adhesion pathway involve MAPK signaling pathways regulating cell survival and the ability to modulate the actin cytoskeleton and cell morphology [50][51][52][53][54]. Kinase interactions involved CBU1724 binding with MAPK10 to modulate cellular stress responses. The remaining effectors targeting the focal adhesion pathway were linked to proteins that regulate actin cytoskeletal remodeling of cell morphology and cell-cell or cell-matrix interactions. The ability to regulate or modulate actin cytoskeleton remodeling in altering cell morphology is also linked Mechanisms of action of Coxiella burnetii effectors inferred from host-pathogen protein interactions to interference in the transitions between phases of the cell cycle, with the polarization state of alveolar macrophages influencing the susceptibility of the host cells to infection [55]. C. burnetii interactions at centrosome microtubules could serve to regulate cell cycle progression by interfering with or altering spindle assembly and, thereby, changing cell polarity. The host targets for these interactions include both regulatory elements, such as ROCK1 and RBL2, as well as the actin-binding protein ANLN.
Pathogen-host interactions that interfered with different components of the immune system involved both a signaling component directly targeting the Fc receptor (FCER1G). Although antibody-mediated immunity to C. burnetii is thought to be Fc receptor independent [56], the ability to directly bind to the receptor may have cascading effects interfering with phagocytosis and cytokine generation. The ubiquitin-proteasome system was affected by both ubiquitin-ligase complex interactions as well as by numerous proteasome sub-unit interactions involving multiple C. burnetii proteins. Modulation of the ubiquitin-proteasome pathway is a feature of the strategy utilized by the closely related pathogen Legionella pneumophila [57,58], which co-opts the host protein degradation system to temporally regulate bacterial effectors in the host cell.
The preferential cellular locations of targeted host proteins (Table 5) also point to important bacterial mechanisms related to endosomal trafficking, vacuole creation, and exosome function [45, 59,60]. They indicated that a large number of pathogen targets associated with extracellular host exosomes form a broad group of C. burnetii-targeted proteins; almost 30% of all protein interactions could be linked to exosomes. Although the targeted host proteins are involved in numerous physiological functions, their main roles are in proteolysis via proteasome interactions, cytoskeletal arrangement via actin regulation, chaperone activity through heat-shock proteins, and oxidative stress responses via thioredoxin interactions. The main function of extracellular exosomes is to enable intercellular host communication, primarily as carriers of immune signals during pathogen infections. For example, Salmonella-infected host cells secrete exosomes enriched in bacterial lipopolysaccharides [61]. The potential ability of C. burnetii to interfere with this process is a novel aspect of the immune evasion by this pathogen.
Proteins linked to endosomal sorting complexes required for transport (ESCRT) also function in vesicle budding, a process used by the uninfected host cell to control membrane abscission during cytokinesis. Intracellular vacuolar pathogens, such as Salmonella, Legionella, and Coxiella, must be able to influence these key host processes to establish their specialized intracellular environments. The interaction set targeting the ESCRT involved 7 pathogen proteins and 11 host proteins associated with 17 total host-pathogen interactions. Of special note is the ER to Golgi protein trafficking host protein TRAPPC8 [62], which potentially interacts with four different C. burnetii effectors; ubiquitin peptidase USP8, which regulates endosome morphology and protein transport [63], and NPC2, which is involved in cholesterol trafficking [64].
The identification of the mitochondrial matrix as a host location targeted by C. burnetii effectors implicated multiple processes that may lead to a more complex mitochondrial phenotype than just inhibition of apoptosis [65][66][67]. Of the 16 host-pathogen protein interactions among 14 host proteins and 11 C. burnetii proteins, 31% were linked to protein synthesis within mitochondria, 19% to fatty acid beta-oxidation, with the rest involving different enzymatic catabolic functions.
The presence of multiple host targets in different pathways points to a possible higher biological or casual organization of bacterial effector interactions. Fig 3 shows the interconnected nature of Coxiella-targeted host pathways according to the number of pathogen-interacting host proteins present in human KEGG pathways. We have broadly arranged and grouped these pathways to highlight RNA processing, protein handling, macromolecular degradation, cellular signaling (including immune-related signaling), and metabolism. Pathway intersections (crosstalk) are indicated by arrows and highlight the potential of effector proteins to affect multiple cellular processes of the host. Although pathway enrichment analysis highlights the presence of multiple targets, each individual interaction could also be critical to the function of the pathway or have physiologically important downstream effects. The biological assumption underlying pathway enrichment analysis is that by affecting multiple targets in a pathway, there is a higher chance that the function of a pathway may be altered. Hence, a bacterium that has become successful at infecting hosts is assumed to have developed a higher propensity to affect multiple proteins in that pathway. Because this may not always be the case, the interpretation of the interaction data should also consider the overall broader set of potentially affected pathways as well as individual sets of effector host-pathogen interactions. The pathway mapping in Fig 3 links major immune and stress signaling pathways to 1) cellular remodeling pathways, 2) different lytic vesicles, proteolysis, and programmed cell death, and 3) metabolic pathways involved in amino acid degradation and energy production.
The aggregate information derived from all host-pathogen interactions represents a spatial and temporal average of these interactions that do not capture infection progression or maintenance per se. What emerges from the interaction profile is a broad pattern of host process interference, which is commensurate with known Coxiella/Legionella infections and strongly Mechanisms of action of Coxiella burnetii effectors inferred from host-pathogen protein interactions emphasizes the ability to affect host organelle creation, endosomal trafficking, and organelle interactions with the parasitophorous vacuole [68][69][70][71][72].

Using Y2H interactions to characterize individual C. burnetii effector function
We used the high-throughput Y2H interactions to identify processes and pathways with a high probability of being affected during the infection. The underlying assumption of this analysisthat multiple protein interactions exert a coordinated effect to re-direct host processes-is based partly on the observation that the bacterium is able to non-specifically infect multiple hosts and cell types and partly on the finding that a large number of effectors are secreted into the host cell. Thus, although some individual interactions may be incorrect, the broader integration patterns of coordinated action should still be evident in these analyses. Conversely, we also used these interactions in conjunction with a smaller pairwise test set to better characterize a set of four effectors for which evidence on their functional significance is lacking.

Complementary pairwise Y2H testing
We carried out high-throughput pathogen-host protein interaction screening using C. burnetii ORFs and human and murine random cDNA libraries. In order to assess the feasibility of using the well-characterized human ORFeome resources to validate the high-throughput Y2H interaction, we selected as subset of C. burnetii-human PPIs and tested them using pairwise Y2H screening. Herein, the Y2H prey clones were generated by cloning the human ORFeome clones to pGADT7g [37], a modified Gateway technology-compatible Y2H prey vector. This prey vector differs in the linker amino acid sequences encoded between the activation domain and the ORFs. Furthermore, the human ORFeome clones do not contain an endogenous stop codon; hence, they encode additional peptide sequences at the C-terminal ends of the ORFs, which may affect their interaction patterns. Previous studies have shown that such differences in either bait or prey vectors produce significantly different interaction data [33,73]. Our expectation was to reproduce approximately 50% of interactions in the pairwise Y2H screening, as minor variations in two-hybrid vectors detect markedly different subsets of interactions in the same interactome space [74].
In this pairwise low-throughput Y2H screen, we individually tested selected interactions involving four relatively uncharacterized C. burnetii effector proteins plus interactions associated with the plasmid CBUA0014 effector examined in the high-throughput screens. Thus, the pairwise testing encompassed 91 host-protein interactions from five C. burnetii proteins-CBU0794, CBU0881, CBU1724, CBU2078, and CBUA0014-which involved 22, 75, 54, 42, and 33 total interactions in the high-throughput screen, respectively. Briefly, of the 226 individual interactions found in the high-throughput screen for these effectors, we constructed Y2H clones and tested 91 or 40% in a pairwise Y2H assay. Of these, 43 (47%) tested positive. There was no correlation between the fraction of interactions that tested positive and the number of interactions in the high-throughput screen, which further suggests that interactions not seen in the pairwise Y2H screening could be due to the intrinsic variation in the vectors and cDNA clone used. The average interaction reproducibility per effector of 47% with a sample standard deviation of 19% was better than, or on par, with other experimental studies [34, [75][76][77]. All individual results are included in S2 Table. Individual C. burnetii effector host-pathogen interactions Table 6 lists known effectors, with known mechanistic associations, which we tested in our high-throughput Y2H screens. Given their importance as pathogen proteins critical for C. burnetii infections, we used the Y2H data to examine both location and functional annotations associated with the targeted host proteins. Table 6 provides their common names, keywords from published literature data, and a summary of the interaction analyses-the number of host interactions, top four candidate cellular locations of interacting host proteins, direct functional evidence from individual host protein interactions, and enriched pathways/locations (Tables 3-5), including host-pathogen interactions where the effector took part. Note that the sum of the fractions of protein locations can exceed one, due to the possible presence of proteins in multiple locations, e.g., if all targeted proteins were present both in the nucleus and in mitochondria, both location fractions would be 100%.
The T4SS effector CBU0781 (ankG) is linked to modulation/prevention of intrinsic host cell apoptosis, located in mitochondria, and subsequently translocated to the nucleus [17,77,78,79]. Of the 40 identified host targets, the four proteins localized to mitochondria (BBOX1, DLAT, ECH1, and OCIAD21) are involved in metabolism. Most (50%) of the targeted host proteins are located in the nucleus, with the remainder primarily located at sites associated with the plasma membrane or different membrane-bound cellular organelles. Of note is the observed Y2H interaction between ankG and KPNA3, the importin subunit α3 of the nuclear pore complex, which is required for the localization of ankG to the nucleus [17]. The ankG-host protein interactions in the nucleus are associated with apoptotic processes involving ABCA1 [80], PARP4 [81], and PGRMC2 [82], as well as interference in cell cycle/ mitotic progression via CDK14 [83], STAG3 [84], RFX5 [85], and BAZ2B [86].
The CBU0937 (cirC) effector was determined to be an immunoreactive Q fever-specific protein [87] and important for intracellular replication. Examination of the host protein targets did not unequivocally identify a virulence mechanism associated with this effector. The primary locations of the host proteins indicated multiple potential sites and mechanisms. Protein targets in the plasma membrane may be related to solute transport and calcium and cationic exchange, whereas those located in mitochondria could be linked to mitochondrial protein synthesis. Targets in the nucleus could be related to stress responses (MAP2K4, PPM1G, and RGS2) and RNA processing (PA2G4 and PAPOLA). The nuclear effector CBU1314 modulates the host transcription response [88], but with scant Y2H interaction data we could only identify nuclear host protein interactions related to focal adhesion (the Rho GTPase activating ARHGAP5 gene) and proteasome subunit interactions involved in the NF-κB immune-response pathway.
CBU1524 (caeA) is linked to intranuclear interactions that prevent or delay apoptosis of the host cell [66,67,89]. Most (65%) targeted host proteins were located in the nucleus, consistent with the finding that caeA is preferentially located in the nucleus [89]. However, substantial fractions were also observed in the plasma membrane (25%) and in the ER/Golgi apparatus (17%). The roles of the targeted nuclear proteins related to different aspects of gene translation, such as splicing (e.g., spliceosome-associated proteins DHX15, SAP18, and SNRPA1), RNA transport, and RNA transcription, as well as ubiquitin-proteasome regulation. Those of the host proteins localized to ER sites were associated with specific ER-associated protein degradation (ERLIN2) and apoptosis (PKD2 [90] and RTN4 [91,92]).
We also investigated the potential interactions of CBU0077, which accumulates at vacuolar membranes and abnormal ER extensions, suggesting that it interferes with vesicular traffic [89]. Recent research tag CBU0077 as a complex-forming effector at the mitochondrial outer membrane during Coxiella infection [93]. Six host proteins interacted with CBU0077, three of which directly related to the reported phenotype (i.e., the junction adhesion molecule AMICA1 [94], the transmembrane-trafficking protein TMED10, and ACADM as part of the mitochondrial fatty acid beta-oxidation pathway. None of these host proteins were located in lysosomes, which are the preferred sites in HeLa cells [30], indicating that the roles of these proteins may vary in different cell types [69]. The presence of effectors on the conserved Coxiella plasmid indicates the importance of their roles. The plasmid effector CBUA0014 (cpeC), an F-box homology protein, localizes to ubiquitin-rich compartments and is hypothesized to play an important role in establishing host infections [95]. Table 6 shows a range of cellular sites where potential host target proteins are preferentially located (nucleus, ER/Golgi apparatus, and plasma membrane), as well as several specific pathways that may be impacted by this effector, including components of the ubiquitin-proteasome system and vacuolar sorting pathway (retromer complexes). Of note are the protein interactions we verified in the complementary pairwise Y2H testing experiment. Thus, cpeC could bind to the ubiquitin-ligase complex component SKP1, which is part of the SKP1-CUL1-F-box protein complex, as corroborated by the observation that ubiquitin colocalizes with this effector [95]. The potential pleiotropic role of cpeC is indicated by the verified Y2H interactions with VPS26A, which is involved in protein sorting of vacuolar pathogens and TMX2 [96] as part of the ESCRT endosomal sorting complex.

Inferring mechanisms of action for relatively uncharacterized effectors
We further selected four additional effectors (CBU0794, CBU0881, CBU1724, and CBU2078) with little or no information on their role during infections for analysis and complementary pairwise Y2H testing, in an attempt to characterize their potential functions. Table 7 provides an overview of these protein interaction analyses, including alternate gene symbols, the success of the complementary testing effort for these proteins, the number of host interactions, the top four cellular locations where interacting host proteins may be located, direct functional evidence from individual host protein interactions, and a list of enriched pathways/locations from Tables 3-5 involving the effector.
The hypothetical protein CBU0794 exhibited interactions with host proteins in the focal adhesion pathway (CAPN2) and with the proteasome (PSMC1), but the relatively low number of interactions and the low number of overall successfully pairwise tested Y2H interactions did not allow us to define a more precise role for this effector.
The potential effector CBU0881 exhibited the second largest number of host-pathogen interactions in our data set involving host pathways/processes, such as endosomal sorting, focal adhesion, the NF-κB immune-response, and protein processing in mitochondria. Complementary Y2H testing confirmed two interactions in mitochondria involving the peptidase PITRM1, which is responsible for degrading transit peptide after their cleavage, and the tRNA-synthetase IARS2, which is involved in mitochondrial protein synthesis. The interaction data also point to extensive interactions of CBU0881 with IST1, ATP6V1E2, CAPN7, and USP8 in the endosomal ESCRT pathway, with only the last interaction retested but not confirmed.
The complementary Y2H testing data confirmed several interactions associated with the centrosome, pointing to a heretofore unrecognized Coxiella-targeted organelle. The highthroughput screens identified a total of 14 host targets located in the centrosome, eight of which interacted with CBU0881. Of five tested interactions, we confirmed three in the pairwise Y2H screen: DCTN6, EFHC, and NEK2. The corresponding functional relationships associated with their centrosome location relate to protein organelle/vesicle transport and spindle morphogenesis (DCTN6 [97]), microtubule-regulation of cell division (EFHC1 [98]), and centrosome separation control and bipolar spindle formation (NEK2 [99,100]). Further indications of the potential pleiotropic role of this effector came from the confirmed interactions associated with nucleus-located host proteins involved in transcription via FAM60A in translational repression and the peptidylprolyl isomerase PPIE in the spliceosome.
CBU1724 interacted with 17 human and 37 mouse proteins with two overlapping interactions ( Table 1). The pathway analyses indicated multiple roles in multiple pathways with selected interactions tested and all confirmed in these pathways (MAPK10, ANLN, AK3, and NPC2). Further evidence of a pleiotropic role for this protein was revealed by the interactions Mechanisms of action of Coxiella burnetii effectors inferred from host-pathogen protein interactions with proteins involved in ubiquitin handling and processing, such as CUL1, MYCBP2, OTUB1, and UBE2V2, with the interaction between CBU1724 and the ubiquitin thioesterase OTUB1 confirmed in the complementary testing.
In addition to the high-throughput screening identification of transthyretin (TTR) as a potential C. burnetii target involved in cholesterol processing [101], complementary testing confirmed two additional host proteins involved in cholesterol handling: the apolipoprotein APOA1BP [102] and the intracellular cholesterol transporter NPC2 [103]. Although C. burnetii infections require cholesterol from the host [64,104], pathogen mechanisms that influence cholesterol uptake and processing in the host are not well characterized. The use of T4SS effectors as potential mediators of this process would thus be critical for maintaining and propagating the infection.
For the final effector, the high-throughput screening pathway analysis and accompanying Y2H pairwise testing data revealed that CBU2078 interacts with host proteins involved in focal adhesion and endosomal sorting. Analysis of the complementary testing data confirmed interactions for four of seven proteins involved in the ubiquitin-proteasome system: RBBP6, RNF38, USP47, and WDR48. These proteins included ubiquitin ligase/proteases (RBBP6, RNF38, and USP47) as well as a protein that stimulates the activity of ubiquitin-specific proteases (WDR48) [105,106].

Y2H characterization of bacterial effector proteins
Large-scale high-throughput Y2H protein interaction screens can yield large amounts of novel and unbiased data. Here, we used this technology to focus on a select set of secreted bacterial effectors in an effort to characterize C. burnetii host-pathogen interactions. We conducted the largest human and murine genome-scale Y2H screening to date, using 33 C. burnetii T4SS effectors to identify and characterize host-pathogen PPIs among this set. The aim of this study was thus to derive general hypotheses on the mechanisms of effector actions mediated by protein interactions. These screens provided partial host interaction data for 25 bacterial effectors distributed among 415 unique pairwise interactions.
Our analyses of these data and the hypotheses derived from studying individual and multiple interactions rely on the underlying protein interactions having biological significance. The nature of protein interactions is complex; high-affinity interactions are not necessarily more biologically relevant than low-affinity interactions because transient interactions with signaling proteins may have important downstream effects not directly captured by PPI data. Many of these low-affinity and transient interactions are strongly dependent on the physiological state of the cell and may vary with the intracellular environment. Similarly, host and pathogen proteins must be simultaneously present in the same cellular compartment and of sufficient quantity for the predicted interaction to be biologically relevant. Thus, although all T4SS effectors are initially released into the cytoplasm of the host cell, their primary targets may be hostcell compartment/organelle specific, and their effects may depend on the ability to efficiently translocate to the correct site of action at the appropriate time.
C. burnetii is capable of colonizing multiple eukaryotic hosts, including different mammalian species and many different cell types. Consequently, the basic infection program executed by the pathogen must exhibit both the generalized and specific features of host-pathway interactions. A corollary of this assumption is that the likelihood of each host-pathogen protein interaction being highly specific, involving strong binary interactions, or being essential for establishing the intracellular infection is small. Thus, we used the high-throughput screening data to identify the biological processes or pathways and cellular locations to which the targeted host proteins belong, in order to characterize aspects of the general impact on the host cell of the tested effectors.

Pathways and processes targeted by C. burnetii
The pathway enrichment analysis of the interactions detected in the Y2H high-throughput screens identified host cellular processes known to occur in Coxiella infections and broadly compatible with both Legionella and Salmonella infection phenotypes. Not surprisingly, many of these related to cellular processes in which direct protein interactions are expected to play a critical role, such as in adapting cell morphology through cytoskeletal re-arrangements, protein processing and trafficking, and vacuolar generation. Similarly, direct protein interactions could strongly affect the frequently observed interactions with components of the ubiquitin-proteasome pathways. The enrichment analyses also indicated targeted host processes involved in amino acid metabolism and cholesterol processing, indicating a potential nutritional component associated with the infection phenotype and T4SS effectors. The interaction data also revealed the exosome to be a preferred location for many of the targeted host proteins. If C. burnetii effectors interfere with the intracellular trafficking of immature exosomes or interact with proteins in exosomes, they would be expected to dampen the immune signaling that occurs between host cells and prevent the host from mounting a more robust immune response.
We also examined the host proteins targeted by specific effectors previously found to alter Coxiella infection phenotypes in transposon mutant studies. We used the interaction data to characterize potential mechanisms of action and identify specific host targets for effectors that could be associated with the observed mutant phenotypes for ankG, cirC, and caeA. In the case of ankG and caeA, we used their nuclear locations in the host cell to focus the analysis on apoptotic control and interference with host gene transcription. In contrast, for other proteins such as CBU1314, the interaction data were limited, suggesting that protein interactions are not involved in their mode of operation.
Finally, we considered the interaction data to derive hypotheses on the potential mechanisms of action of four individual effectors for which phenotypic information is sparse. In three cases, the joint interaction data supported mechanisms that were broadly similar to those of the other effectors, such as involvement in ESCRT processes, ubiquitin handling, and transcriptional control; at the same time, they also pointed to specific processes not previously associated with T4SS effectors, such as cholesterol transport (CBU1724) and centrosome interactions (CBU0881). As characterizing novel effectors is often referred to as "the hard part" [2], careful in vitro and in vivo phenotypic studies of deletion mutants will be needed to acquire more precise information on their specific roles.
The present study delineates the largest collection thus far of C. burnetii effector protein interactions derived from whole-genome human and murine host protein libraries. This data set represents a rich source of information on secreted pathogen effector proteins and their interactions with host proteins. As such, these interactions contain critical information on how the pathogen establishes and maintains a successful intracellular infection.
Supporting information S1 Table. Excluded host proteins. The table provides host proteins known to be indiscriminate binders and eliminated from our yeast two-hybrid high-throughput screen. (XLSX) S2 Table. Host-pathogen interaction data. The table provides all pairwise interactions detected in the yeast two-hybrid (Y2H) high-throughput screen from the human libraries and the corresponding murine orthologous interactions as well as the pairwise interactions from the complementary pairwise Y2H testing. (DOCX)