Bats are key hosts in the radiation of mammal-associated Bartonella bacteria

Abstract

Bats are notorious reservoirs of several zoonotic diseases and may be uniquely tolerant of infection among mammals. Broad sampling has revealed the importance of bats in the diversification and spread of viruses and eukaryotes to other animal hosts. Vector-borne bacteria of the genus Bartonella are prevalent and diverse in mammals globally and recent surveys have revealed numerous Bartonella lineages in bats. We assembled a sequence database of Bartonella strains, consisting of nine genetic loci from 209 previously characterized Bartonella lineages and 121 new cultured isolates from bats, and used these data to perform a comprehensive phylogenetic analysis of the Bartonella genus. This analysis included estimation of divergence dates using a molecular clock and ancestral reconstruction of host associations and geography. We estimate that Bartonella began infecting mammals 62 million years ago near the Cretaceous-Paleogene boundary. Additionally, the radiation of particular Bartonella clades correlate strongly to the timing of diversification and biogeography of mammalian hosts. Bats were inferred to be the ancestral hosts of all mammal-associated Bartonella and appear to be responsible for the early geographic expansion of the genus. We conclude that bats have had a deep influence on the evolutionary radiation of Bartonella bacteria and their spread to other mammalian orders. These results support a ‘bat seeding’ hypothesis that could explain similar evolutionary patterns in other mammalian parasite taxa. Application of such phylogenetic tools as we have used to other taxa may reveal the general importance of bats in the ancient diversification of mammalian parasites.

1. Introduction

A central part of the work done by disease ecologists is to understand the host range of infectious agents. However, host ranges must be understood in a coevolutionary context, specifically how agents have adapted to and diversified in hosts over time. Only by considering both ecological and evolutionary context can we understand how agents come to infect and adapt to new hosts. While cophylogeny is a common tool for studying the codiversification of hosts and parasites, few studies have examined the relative timing of the diversification of parasite lineages in parallel with that of hosts (Roth et al., 2019; Zhu et al., 2015).

The genus Bartonella is an excellent study system for disease ecology and evolution because it is common and diverse in many mammalian hosts (Kosoy, 2010). These alphaproteobacteria are facultative intracellular pathogens that can cause persistent, hemotropic infections in their hosts. Transmission between hosts occurs through a variety of hematophagous arthropod vectors, wherein bartonellae colonize the midgut and are then shed in arthropod feces (Harms and Dehio, 2012). Clades of Bartonella species tend to be host-specific (Vayssier-Taussat et al., 2009), so it could be hypothesized that the genus diversified along with its mammalian hosts millions of years ago. However, there have been few comprehensive phylogenies of this genus and limited research on the influence of particular host groups on Bartonella evolution.

Bats are a group of special interest because they have traits that are amenable to parasite transmission, including their global distribution, ability to fly, seasonal migration, dense aggregations and high sociality in some species, long life spans, and the use of torpor and hibernation (Calisher et al., 2006). There is also evidence that chiropteran immune systems are highly tolerant of infections, especially of viruses (Banerjee et al., 2020). Thus, their role as reservoirs for Bartonella bacteria may be especially influential among mammals. Bats are also an ancient clade of mammals (Shi and Rabosky, 2015), providing ample time for diversification of bacterial parasites and transitions from bats to other mammals. Research has concluded that bats are potentially ancestral hosts that influenced the diversification and spread of coronaviruses (Cui et al., 2019; Latinne et al., 2020), lyssaviruses (Badrane and Tordo, 2001), paramyxoviruses (Drexler et al., 2012), trypanosomes (Cĺement et al., 2020; Espinosa-Álvarez et al., 2018), and haemosporidia (Galen et al., 2018; Lutz et al., 2016) among other mammalian orders. Drawing from Hamilton et al. (2012), who developed the ‘bat-seeding’ hypothesis to explain the geographic and host distribution of Trypanosoma lineages related to the agent of Chagas disease, T. cruzi, we hypothesize that bats may have also been influential in the ancient diversification and spread of Bartonella bacteria.

Successful amplification of Bartonella DNA from recent fossils points to a prolonged history of Bartonella infection in some hosts, such as humans and domestic cats (Fournier et al., 2015; Mai et al., 2020). However, it is unlikely that DNA could be successfully amplified from more ancient fossils to test hypotheses about the origin of bartonellae in mammals. Instead, a molecular clock approach can be used to estimate the rate at which substitutions accumulate in Bartonella DNA and then extrapolate divergence dates of clades. Considering new research has shown that mammal-associated bartonellae evolved from arthropod symbionts (Segers et al., 2017), we rely on a molecular clock for the 16S ribosomal RNA (rRNA) gene based on sequence divergence data from bacterial symbionts of arthropod hosts separated for millions of years (Kuo and Ochman, 2009). We perform a multi-locus analysis using a comprehensive database of Bartonella strains, including a greater number of loci than a recent time tree analysis (Frank et al., 2018) and a previous global analysis of Bartonella lineages in bats (Corduneanu et al., 2018), and broader taxon sampling than previous genomic analyses (Harms et al., 2017). Many new Bartonella strains have recently been discovered in bats (Corduneanu et al., 2018; McKee et al., 2016; Nabeshima et al., 2020), so we have included 121 novel isolates from bats in this study to amend current delineation of Bartonella clades (Harms et al., 2017) and to determine the influence of bats on the diversification and spread of Bartonella bacteria to other mammalian orders.

Using this molecular clock approach, we extrapolated when the genus Bartonella diversified and compared the timing of Bartonella clade diversification along with their hosts. We hypothesized that mammal-infecting bartonellae evolved with their hosts starting in the late Cretaceous or early Paleogene when many eutherian and metatherian taxa diversified (Phillips, 2016). We expected to see clustering of Bartonella lineages by host orders and correlation between diversification dates of hosts and Bartonella clades. Using ancestral state reconstruction and network analysis, we discerned which orders of mammals were highly influential in the diversification and spread of Bartonella to other host orders and geographic regions. We predicted that the speciose orders of bats (Chiroptera) and rodents (Rodentia) are important in the historical expansion of the Bartonella genus, but that bats may have had a more profound influence in this process because of their ability to fly and quickly disperse over wide areas. This study provides a more complete understanding of Bartonella evolution and biogeography and the role of bats as important hosts of pathogens through a suite of phylogenetic methods that can be adapted to understand these processes in other host-specific parasites and symbionts. Such investigations could lead to a deeper evolutionary understanding of symbiosis and parasitism and the identification of key host groups in the diversification and spread of these organisms.

2. Materials and methods

2.1. Molecular data collection

Accurate phylogenetic analysis requires a balance between adequate taxon sampling to represent lineage diversity and sufficient genetic information (i.e., sequence length or number of sequenced loci) to produce a statistically robust phylogeny (i.e., with high branch support across the tree). Limitation in either one of these dimensions may lead to erroneous conclusions about the evolutionary history of organisms (Heath et al., 2008). However, constraints on the availability of data or computational resources may require some compromise of these dimensions and still produce useful phylogenetic inferences. In an attempt to balance the need for adequate taxon sampling and sequence data, we assembled a database of Bartonella sequences from published genomes on GenBank, previous studies using multi-locus sequence analysis (MLSA), and archived cultures from bats. MLSA is a well-established method in microbiology for inferring the evolutionary history of pathogenic bacteria that can be more accurate than single-locus analyses (Ahmed et al., 2006; Margos et al., 2009; Margos et al., 2008; Zhu et al., 2005). We targeted nine genetic markers (Supplementary Table S1) commonly used for Bartonella detection and phylogenetic analysis (Kosoy et al., 2018; La Scola et al., 2003). Data from MLSA studies and genomes published as of 2018 were collected from GenBank via accession numbers or strain numbers from 74 studies (Supplementary Dataset 1), including recent publications that have isolated bartonellae or related bacterial symbionts in arthropods and past studies characterizing bat-associated Bartonella strains from Asia, Africa, and North America. We excluded any strains that were noted in the studies as showing evidence of homologous recombination between Bartonella species to prevent issues with incomplete lineage sorting in phylogenetic analysis. Additional molecular data collection of Bartonella isolates from bats included a subset of cultures archived in our laboratory from previous studies in Africa, North and South America, Europe, and Asia that have been partially characterized at some of the targeted loci, as well as new cultures from bats sampled from Nigeria in 2010 and Guatemala in 2010, 2014, and 2015. The data combined from bat-associated Bartonella strains cover 50 species from 11/21 extant chiropteran families (Simmons and Cirranello, 2020). Details on sequencing of bat-associated Bartonella isolates, alignment, data cleaning, and validation can be found in the Supplementary Material. The final database (including GenBank accession numbers) contained sequence data from 332 taxa: 209 Bartonella reference strains from genomes and MLSA studies, 121 bat-associated isolates from our laboratory archive, the ant symbiont Candidatus Tokpelaia hoelldoblerii, and the outgroup Brucella abortus (Supplementary Datasets 1 and 2). The taxonomic and geographic diversity of bat species represented with Bartonella lineages in our dataset is similar to two recent global analyses of Bartonella diversity in bats (Corduneanu et al., 2018; Frank et al., 2018) and generally reflects the sampling effort of Bartonella studies in bats performed to date (Supplementary Figs. S9 and S10).

2.2. Phylogenetic analysis

Bayesian phylogenetic analysis was performed using BEAST v1.8.4 (Drummond and Rambaut, 2007) via the CyberInfrastructure for Phylogenetic RESearch (CIPRES) Science Gateway portal v3.3 (Miller et al., 2010). The nine loci were partitioned separately using GTR + I + G sequence evolution models, estimated base frequencies, four gamma rate categories, an uncorrelated relaxed clock with an exponential distribution of clock rates along branches for each locus, and a birth-death speciation model with incomplete sampling (Stadler, 2009). Brucella abortus was set as the outgroup in all analyses. To determine a clock prior for the 16S rRNA locus, we analyzed published 16S rRNA sequence divergence and host divergence times for bacterial symbionts of arthropods (Kuo and Ochman, 2009). A linear regression model was fit to the data in R (R Core Team, 2020) and a lognormal prior was estimated by moment matching to the normal distribution for the fitted mean and standard error of the slope (Supplementary Fig. S5). The prior distribution for the exponential clock rate for 16S rRNA was set to this lognormal distribution while prior distributions for the exponential clocks of the remaining eight loci were set to an approximate reference prior for continuous-time Markov chain rates (Ferreira and Suchard, 2008) due to a lack of relevant prior information on evolutionary rates for the other sequenced loci. Thus, the 16S rRNA clock acted as a strong prior and the rates for the other eight loci were estimated relative to the 16S rRNA rate. This approach allowed for external validation of Bartonella diversification events based on host diversification dates without explicitly using host diversification dates as calibration points for the parasite tree. Extensive testing using alternative substitution (with or without codon partitioning), clock, and tree models and subsets of genetic data determined that model choice or the exclusion of the ITS locus had little influence on tree topology and estimated divergence dates (Supplementary Table S5). Additional details regarding model priors and run settings can be found in the Supplementary Material.

2.3. Ancestral state reconstruction

In addition to divergence time estimation, we performed ancestral state reconstruction in BEAST. We assigned discrete traits to each tip based on the taxonomic order of the host and the geographic ecozone (Olson et al., 2001 that includes the majority of the host’s geographic range. The association of some Bartonella lineages with arthropods and not mammals are justified in the Supplementary Material. Ancestral state reconstruction was performed using a symmetrical rate model to reduce the number of state transitions that needed to be inferred.

2.4. Tip-association tests

We performed tip-association tests using the Bayesian Tip-association Significance testing (BaTS) program v1 to assess the clustering of traits along tips of the phylogenetic tree (Parker et al., 2008). We performed four sets of simulations using the same assignments of host orders and geographic ecozones used in the ancestral state reconstruction above. The two sets of traits were simulated on 1000 posterior sampled trees from the final BEAST run and on the single maximum likelihood (ML) tree. Clustering of traits was measured by the association index (AI) and parsimony score (PS), producing a distribution for the 1000 Bayesian trees and a single value for the ML tree. Null distributions for these measures were generated using 100 randomizations of traits onto tips of the trees. The significance of clustering was evaluated based on the overlap between observed values or distributions of AI and PS and their null distributions. For both measures, small values indicate a stronger phylogeny-trait association (Parker et al., 2008).

2.5. Host clade definitions and divergence dates

We defined host-associated Bartonella clades a posteriori based on high posterior support (>0.9) and clustering by host orders from the ancestral state reconstruction (Fig. 1A). Previous analyses of Bartonella host associations have shown that host-switching is common (Lei and Olival, 2014), so a calibration approach that assumes strict cospeciation across the tree would not accurately reflect the evolutionary history of these bacteria. However, Bartonella lineages are broadly host-specific within orders (Frank et al., 2018) and host-switching is more frequent between closely related hosts (McKee et al., 2019). We defined 15 host-associated Bartonella clades (Supplementary Tables S5–S6) at relevant taxonomic scales below the order level to test the hypothesis that Bartonella lineages diversified with their hosts while accounting for frequent host-switching that could occur within a host clade. We collated divergence dates for the most recent common ancestor uniting the host taxa of interest within each clade from available studies in the TimeTree database (http://timetree.org/), summarized by the estimated mean, 95% confidence intervals, and range of dates across studies (Kumar et al., 2017). We then correlated these mean host divergence dates with our estimated median divergence date of the associated Bartonella clade (Supplementary Table S7). A significant linear fit between these dates would support the hypothesis that Bartonella clades diversified with their hosts after colonization. To validate measurement of the divergence times for mammal-associated Bartonella clades with the ultrametric tree produced in BEAST, we also generated a calibrated timed phylogeny with the ML tree. Using the RelTime relative rate framework (Tamura et al., 2012) within MEGA v10.0.5 (Kumar et al., 2018) we generated a timed phylogeny using host clade divergence dates from TimeTree (Supplementary Table S7). We used confidence intervals (or ranges in the case of clade J) for the 15 host clade divergence dates as minimum and maximum divergence dates in RelTime. The program then calculated divergence dates on the tree using a maximum likelihood approach (Tamura et al., 2012), producing mean estimates and 95% confidence intervals for clade dates that we could compare with the eubartonellae date estimated in BEAST. This analysis can confirm that divergence date estimation is robust to different approaches by comparing a calibration-based method on an existing tree to a method that relies on relaxed clock priors during tree estimation.

Fig. 1.

Evolution of Bartonella lineages, host associations, and geographic origins. Timed maximum clade credibility tree of Bartonella lineages. Tips are collapsed into clades of related Bartonella species and strains (Supplementary Tables S5 and S6); the number of tips in each clade is shown in brackets. Posterior probabilities (PP) for nodes are indicated by the size of circles; ancient nodes had strong support (PP = 1), unless otherwise labeled. Branch lengths are in millions of years. Ancestral state reconstruction of (A) host order and (B) ecozone transitions was performed during the estimation of diversification times. Branches are colored according to their most probable (PP > 0.5) host order or ecozone states, with host or ecozone probability shown by the color of circles at each node. The Cretaceous-Paleogene extinction event is drawn as a gray line at 66 million years ago.

2.6. Stochastic character mapping and network analysis

To determine the inferred ancestral host order and ecozone of mammal-infecting eubartonellae (Bartonella species excluding B. tamiae and B. apis), we initially inspected the results of the ancestral state reconstruction on the maximum clade credibility (MCC) tree. Specifically, we inspected the posterior support for the node and the posterior probability of the host order and ecozone at the node across all posterior trees. However, due to the large number of Bartonella lineages associated with Chiroptera in the database (n = 160) relative to those in other diverse orders (Rodentia, 87; Artiodactyla, 32; Carnivora, 21), we tested the influence of this sampling bias on uncertainty about ancestral states using stochastic character mapping of host orders and ecozones onto trees (Huelsenbeck et al., 2003). We wrote a custom R function to resample tips from the phylogenetic tree and perform stochastic character mapping on the pruned tree using the packages ape and phytools (Paradis et al., 2004; Revell, 2012) assuming an equal-rates model. The function ran 100 mapping simulations on each pruned tree and calculated the probability that Chiroptera and Palearctic were the inferred host order and ecozone at the node uniting eubartonellae. These states were chosen based on initial reconstructions from BEAST indicating them as ancestral traits. We performed this simulation using three resampling schemes: equalizing the number of tips associated with bats and rodents (n = 87), equalizing tips associated with bats, rodents, and artiodactyls (n = 32), and equalizing tips associated with bats, rodents, artiodactyls, and carnivores (n = 21). Resampling schemes were run with 100 resampling iterations on the MCC tree and 10 resampling steps on 10 randomly sampled posterior trees. We summarized the resulting probability distributions by the mean and interquartile range (Supplementary Table S11). We further assessed the nature of transitions between hosts and ecozones by performing additional stochastic character mapping simulations on posterior trees followed by network analysis of state transitions. Host orders and ecozones were simulated with phytools over 1000 posterior sampled trees with an equal-rates model. The number of state transitions were then summarized over all 1000 simulations by the median and 95% credible intervals, ignoring state transitions with a median of zero (Supplementary Table S12). Separate host order and ecozone networks were then built from these median transitions, and node-level properties including degree, out-degree, and betweenness centrality were calculated using the R package igraph (Csárdi and Nepusz, 2006).

3. Results

3.1. Phylogeny and age estimation of the Bartonella genus

Using molecular data from nine genetic loci sequenced from 331 Bartonella lineages (Supplementary Table S1), we produced a Bayesian phylogeny with high statistical support for branches across the tree (Fig. 1; Supplementary Fig. S8) that confirmed the monophyly of Bartonella clades identified in past studies (Harms et al., 2017). These included a clade containing rodent-associated B. elizabethae, B. grahamii, B. tribocorum, and B. rattimassiliensis (clade H); a clade containing cat-associated B. henselae and B. koehlerae (clade F), B. quintana, and B. washoensis (clade E); and all three B. vinsonii subspecies (clade K). However, our approach has substantially altered the order of the deep branches within the phylogeny, including the delineation of five distinct Bartonella clades restricted to bats. The relaxed molecular clock estimated over the tree inferred that the mammal-infecting eubartonellae (excluding B. apis and B. tamiae) began diversifying 62 million years ago (mya; 95% HPD: 40–90), near the Cretaceous-Paleogene boundary 66 mya (Fig. 1; Supplementary Fig. S6). Additional details regarding revisions to the Bartonella tree topology and clock rate estimates for sequenced loci can be found in the Supplementary Material.

3.2. Diversification of bartonellae with hosts

Following the hypothesis that the Bartonella genus radiated with its mammal hosts, we performed tip-association tests to analyze clustering by host taxonomy and geographic origin along the tips of the tree. Simulations using 1000 posterior sampled trees showed significant clustering of host orders and geographic ecozones across the phylogeny according to association indices (AI) and parsimony scores (PS). Observed distributions for both measures did not overlap their respective null distributions based on random associations of traits to tips (Supplementary Table S10). Host orders had smaller values for AI and PS than geographic origin, indicating a stronger phylogeny-trait association with host taxonomy than geographic origin. This phylogeny-trait association with host taxonomy is illustrated in Fig. 1A through strong support for monophyletic groups associated with host orders.

We clarified this association with host taxonomy by describing 15 Bartonella clades (Supplementary Tables S5 and S6) predominantly associated with marsupials (B), ruminants (C), carnivores (F), rodents (E, H, I, J, K, M, O), and bats (A, D, G, L, N; Fig. 1A). We then compared divergence dates of each Bartonella clade with divergence dates of the associated hosts within each clade (Supplementary Table S7) collated from TimeTree (Kumar et al., 2017). We found a strong correlation between Bartonella and host clade divergence times (R² = 0.72, F = 36.4, P < 0.0001). However, most (13/15) Bartonella clades were younger than their associated host clades; on average, the age of Bartonella clades was 76% that of their associated host clades (Fig. 2).

Fig. 2.

Comparison of divergence dates between Bartonella clades and associated host mammal clades. (A) Divergence times and intervals for Bartonella (in orange) and host clades (in purple). Black points show the median estimates and thin bars show divergence date ranges. Thick bars for mammal clades are the 95% confidence intervals estimated from TimeTree and the same bars for Bartonella clades are the 95% HPD intervals. (B) Correlation of median divergence dates between host and Bartonella clades, with clade identifiers shown as points. The solid green line indicates the best linear fit through the points and the dashed gray line shows the 1:1 line if host and Bartonella divergence dates were equal.

The Bayesian tree used in these analyses was similar to a maximum likelihood (ML) tree produced from concatenated sequences of all nine loci, with only minor differences in topology for some internal branches and external branches with low bootstrap support (Supplementary Fig. S7). Tip-association tests using the ML tree showed similar results to the Bayesian tree (Supplementary Table S10). Using confidence intervals for host clade divergence dates provided from TimeTree as calibration dates on the ML tree within the RelTime relative rate framework (Tamura et al., 2018), we estimated the origin of mammal-infecting eubartonellae at 66.3 mya (95% CI: 63.5–69.1). This separate analysis validates the Bayesian relaxed clock estimate (Supplementary Table S5) and further supports the inference that Bartonella began diversifying with mammals near the Cretaceous-Paleogene boundary.

3.3. Influence of host groups and geography on Bartonella evolution

Bats appear to be highly influential in the diversification and spread of Bartonella geographically and to other host orders. Bat-associated clades (A, D, G, L, N) are broadly distributed across the tree and form external branches to clades associated with other mammalian orders (Fig. 1A). This contrasts with clades associated with marsupials, ruminants, carnivores, and rodents, which are less dispersed on the tree and stem from more internal branches. Based on ancestral state analysis using host orders as states, bats were inferred to be the ancestral host of all mammal-infecting eubartonellae with a posterior probability of 0.99. Due to the large number of bat-associated lineages in the database (n = 160), this inference of the ancestral host may have been biased towards bats. Yet in all resampling scenarios, the median posterior probability that bats are the ancestral hosts of mammal-infecting eubartonellae exceeded 0.9 (Supplementary Table S11).

In addition to ancestral host associations, we also inferred the ancestral biogeography of Bartonella clades and where host transitions may have occurred. We performed ancestral state reconstruction of ecozones based on the current geographical distribution of the host of each Bartonella lineage. The geographic origin of eubartonellae was inferred to be in the Palearctic (Fig. 1B) with a posterior probability of 0.99. However, the inference of the geographic origin of eubartonellae is less certain when host sampling bias was accounted for in the stochastic character mapping analysis. The median posterior probability for a Palearctic origin of eubartonellae ranged from 0.63 to 0.77 across all resampling scenarios (Supplementary Table S11).

We explored the influence of particular hosts on the spread of Bartonella among mammalian orders and across ecozones using stochastic character mapping and network analysis. After mapping the number of host and ecozone transitions across 1000 posterior sampled trees, we built a network consisting of host orders and ecozones as nodes and the median number of transitions between nodes as edges (Fig. 3; Supplementary Table S12). In general, the ecozone network was more highly connected than the host network (Fig. 3). The higher number of connections in the ecozone network corresponds with the results of the tip-association tests (Supplementary Table S10), which showed that clustering of traits was stronger for host taxonomy than geographic origin. That is, the high frequency of transitions between ecozones leads to lower levels of geographical clustering on the tree.

Fig. 3.

Transition network for (A) host orders and (B) ecozones across the Bartonella phylogeny. Edges connecting nodes are the median number of transitions between host and ecozone states based on stochastic character mapping on 1000 posterior sampled trees. Edge widths are proportional to the median number of transitions. Edges with a median of zero transitions are not shown. Transitions between the outgroup (Brucella abortus) and between mammalian orders and arthropods have been removed for clarity. All transition counts with a median above zero are shown in Supplementary Table S9.

Examining the network properties of the nodes, we found that certain host orders were influential in the spread of Bartonella among host orders (Supplementary Table S13). In particular, we considered degree (the number of edges connected to a node), out-degree (the number of edges originating from a node), and betweenness (the number of shortest paths that connect any two nodes in the network that pass through the node in question) because these measures describe how each node serves as a source of Bartonella to other nodes. Bats and rodents were a source to other mammalian orders (Fig. 3A), with the highest degree and out-degree of all host orders and high betweenness (Supplementary Table S13). Transitions between ecozones show that the historical movement of Bartonella by hosts led to the present global distribution of these bacteria (Fig. 1B) through bidirectional exchange (Fig. 3B). Among the ecozones, Palearctic and Indo-Malayan ecozones showed the highest degree, out-degree, and betweenness (Fig. 1B; Supplementary Fig. S8B).

4. Discussion

Bartonella is a broadly distributed bacterial genus associated with mammals and arthropod vectors globally. Patterns of host-specificity and phylogenetic diversity in this genus reflect general trends in other zoonotic pathogens. Thus, Bartonella serves as a model system for understanding the evolution and ecology of zoonotic agents. Specifically, this system could inform theory about how agents adapt to and diversify in hosts over time and the ecological conditions that lead to accidental infections and host-switching. Using a multi-faceted analytical approach, this study answered several key questions about the evolution of Bartonella bacteria.

First, we confirmed previous studies (Frank et al., 2018; Segers et al., 2017) showing that the genus first evolved as a symbiont of arthropods, represented by the species B. apis, B. tamiae and the ant symbiont Candidatus Tokpelaia hoelldoblerii, before transitioning to a parasitic life-style in mammals. Second, we found that the Bartonella genus began diversifying with mammals around the Cretaceous-Paleogene boundary 66 mya. Many crown metatherian and eutherian clades began diversifying around this time (Phillips, 2016), including the diverse placental orders Chiroptera, Artiodactyla, Carnivora, Rodentia, and Primates, suggesting that Bartonella diversification is tightly linked with the radiation of its mammalian hosts during the Paleogene. This inference was robust to two separate methods of divergence date estimation: the relaxed molecular clock approach in the Bayesian analysis and the relative rates framework using host clade divergence dates on the maximum likelihood tree. Fourth, the use of ancestral state reconstruction or stochastic character mapping of host traits paired with network analysis is a nascent approach in the study of infectious agents that can provide additional insights from phylogenies (Clément et al., 2020; Faria, 2013; Kellner et al., 2018; Latinne et al., 2020). These analyses demonstrated that bats have been key to both the origin and spread of Bartonella among other mammals and geographic regions, while rodents were responsible for additional spread. In further support of our finding that bats are the inferred ancestral host of Bartonella in mammals, the diversification of mammal-infecting eubartonellae started almost exactly when bats began their evolutionary radiation around 62 mya (95% CI: 59–64, range: 51.9–74.9) according to compiled studies from TimeTree (Kumar et al., 2017). We also found that the geographic origin of eubartonellae appears to be in the Palearctic, although with more uncertainty than the inference about a bat origin for eubartonellae. This finding fits with the classification of bats within the clade Laurasiatheria and previous reconstructions of chiropteran biogeography which found that extant bats may have originated in Eurasia (Teeling et al., 2005).

This work elucidates key aspects of the ecology and evolution of Bartonella, yet there are several avenues of research to be explored in future studies. One necessity is to thoroughly catalog Bartonella diversity. While description of Bartonella species was slow through the 20th century, the advent of genetic sequencing has brought about an explosion of Bartonella diversity with over 40 named and likely many other unnamed species. Our phylogenetic analysis used a comprehensive sequence database, including broad taxon sampling of Bartonella lineages characterized from 10 mammalian orders. These data, along with a relaxed clock approach, have reshaped the Bartonella phylogeny, defining five new clades of bat-associated Bartonella lineages and reorganizing the relationships of deeply branching clades. However, the findings of this analysis only reflect the diversity of Bartonella lineages discovered in bats and other mammalian groups sampled to date. Attempts to culture and characterize novel Bartonella lineages from undersampled bat families, mammalian orders, other potential vertebrate hosts (e.g., xenarthrans and birds (Calchi et al., 2020; Williams and Dittmar, 2020)), and arthropods (Bisch et al., 2018) are needed to further improve taxon sampling. This continued work will undoubtedly reshape the Bartonella tree further and may lead to new conclusions about ancient associations with hosts.

Our results also provide context to the biological changes that are associated with the shift of Bartonella bacteria from an arthropod symbiont to a mammal parasite. Our phylogeny reaffirms work demonstrating this shift (Frank et al., 2018; Segers et al., 2017) and provides an estimated time for when it occurred, suggesting that an existing bacterial population colonized a new niche in mammals shortly after their emergence as potential hosts. Some of the molecular machinery that could have facilitated this colonization was already present in arthropod-associated Bartonella lineages and other Rhizobiales bacteria (Segers et al., 2017). The majority of virulence factors important for host interaction or establishment of intracellular infection are shared across Bartonellaceae, suggesting some latent potential for infecting vertebrates even in arthropod-associated lineages. However, the evolutionary radiation of eubartonellae is associated with a number of other important molecular innovations, including the loss of flagella and acquisition of trw and virB type IV secretion systems (T4SS) (Engel et al., 2011; Harms et al., 2017; Segers et al., 2017; Wagner and Dehio, 2019). Secretion systems have only been detected and characterized in a few Bartonella species across the phylogeny, so our revision of Bartonella tree topology highlights a need for future work regarding the machinery (e. g., flagella, T4SS) shared between bat-associated lineages and their relatives.

Given that current mammal-associated bartonellae are vectored by blood-feeding arthropods and ancestral bartonellae were likely arthropod symbionts (Segers et al., 2017), it is probable that early adaptation to blood-feeding arthropods facilitated the colonization of the mammalian bloodstream. Hematophagous arthropods frequently harbor endosymbionts to cope with their nutritionally deficient diet (Husnik, 2018), so ancient (and possibly some extant) bartonellae may have had beneficial relationships with arthropod hosts. The switch from symbiont to mammal parasite could then have occurred early in the evolution of mammals. There is evidence that ancestors of extant mammalian ectoparasites implicated as Bartonella vectors (Leulmi et al., 2015; McKee et al., 2018; Tsai et al., 2011) were already present by the end of the Cretaceous, including sand flies (Akhoundi et al., 2016), fleas (Zhu et al., 2015), sucking lice (Light et al., 2010), bed bugs and bat bugs (Roth et al., 2019), and hippoboscoid flies (de Moya, 2019). The bat-associated families of hippoboscoid flies (bat flies, Nycteribiidae and Streblidae) and the family of fleas specific to bats (Ischnopsyllidae) diverged around 45–55 mya, shortly after the divergence of their bat hosts (de Moya, 2019; Dittmar et al., 2015; Zhu et al., 2015). Based on available evidence, the colonization of mammals by Bartonella bacteria may have occurred via a hematophagous vector, possibly parasitizing early bats. An ancestral relationship with bats is supported by recent detection of B. tamiae in bat flies and bat spleens (Bai et al., 2018; Leulmi et al., 2016), suggesting that this species can opportunistically colonize bats from arthropods even today. The initial transmission may have occurred through contamination of skin with arthropod feces containing bacteria, direct consumption of an infected arthropod (Ramanantsalama et al., 2018), or some other unknown route (Harms and Dehio, 2012). Once inside the host, the existing ability of bartonellae to invade host cells may have led to proliferation of bacteria in the blood. Additional studies that isolate Bartonella lineages in arthropods and confirm potential transmission routes between mammal hosts and arthropod vectors will clarify the evolution of host-vector-Bartonella relationships.

As apparent in Figs. 1 and 3, the evolutionary history of Bartonella has involved several host-switching events. Thus, calibrating divergence dates by relying on codivergence between host taxa would poorly reflect this history. Instead, we initially avoided a calibration approach in favor of using a relaxed clock prior, then validated estimated divergence dates based on 15 radiation events within particular bat, rodent, ruminant, and marsupial host taxa. The Bartonella divergence dates correlate strongly with the host divergence dates, although with a widespread delay in the colonization of Bartonella within a clade (Fig. 2). While it is possible that this delay in Bartonella colonization is associated with the divergence date estimation approach and bacteria diverged immediately along with their hosts, we suspect the delay reflects some biological reality. According to Manter’s rules (Manter, 1966; Manter, 1955), parasites evolve more slowly than their hosts due to the relatively uniform environments they experience within a host. This slow evolution may help to explain rampant Bartonella host-switching between related hosts in the tree, since from a parasite’s perspective the intracellular environments of phylogenetically similar hosts are unlikely to have significantly changed. Despite these inherent delays, the clustering of Bartonella strains with host orders and particular clades within those orders along with the correlation of divergence times strongly suggest a shared evolutionary history between Bartonella strains and their hosts, although a more complicated one than simple cospeciation.

Beyond patterns of codiversification, it is clear from this study that Bartonella evolution has been shaped by certain hosts, particularly rodents and bats. As the two most speciose groups of mammals, they could be expected to host diverse parasites according to Eichler’s rule (Eichler, 1942), which predicts positive covariance between host and parasite diversity. While more studies will need to be done to explicitly test patterns of host and Bartonella diversity while accounting for sampling biases, it is clear from the network analysis that rodents and bats are important sources of bartonellae to other hosts (Fig. 3). As abundant taxa within ecosystems, rodents and bats could act as targets for both generalist and specialist ectoparasites. While endemic Bartonella infections are likely maintained by transmission by specialist ectoparasite vectors, generalist vectors could target the most abundant species in the community (e.g., rodents or bats) and occasionally infest alternative hosts, resulting in opportunities for accidental Bartonella infections in phylogenetically distant hosts over evolutionary time (McKee et al., 2019). Reconstructing some of these ancient host-switching dynamics would require knowledge of ancestral ectoparasite associations and the interactions of hosts and their ectoparasites within communities.

Finally, bats were identified as the most probable ancestral host of eubartonellae in mammals even after accounting for sampling bias in the database. The fact that bats can fly would have hypothetically increased their dispersal ability during their early diversification. This is exemplified by numerous long-distance colonization events: from mainland Africa to Madagascar by seven different extant bat families, including the endemic Myzopodidae; from Australia to New Zealand by the family Mystacinidae; and from mainland North America to Hawaii by Lasiurus cinereus (Eick et al., 2005). The dispersal of bats to distant landmasses during the early diversification of extant mammals could have played a role in the importance of bats as sources of Bartonella infection to other hosts. We also note that bats appear to be highly tolerant of infections, especially of intracellular bacteria and viruses (Brook and Dobson, 2015), showing few signs of disease and unique immune responses compared to other mammals (Ahn et al., 2019; Banerjee et al., 2020; Hayman, 2019; Schountz et al., 2017). Such patterns in extant bats may have ancient origins linked with their ability to fly (Zhang et al., 2013) and thus bats may have been ideal hosts for the early colonization of mammals by arthropod-borne bartonellae.

The importance of bats in the evolutionary diversification of mammal parasites has been discussed by other authors working in distinct systems. One of these groups are the Trypanosoma parasites that include T. cruzi, the agent of Chagas disease. Observing the broad distribution of bat-associated clades in the growing diversity of trypanosomes, Hamilton and others hypothesized that bats may have been highly influential in the geographic spread of the T. cruzi clade and host-switching to other mammals (Hamilton et al., 2012). This ‘bat-seeding’ hypothesis has continued to gain support since it was proposed with the discovery of diverse lineages in the T. cruzi clade in bats globally (Clément et al., 2020; Espinosa-Alvarez et al., 2018). Similar patterns have been noted in malarial parasites (Haemosporida), wherein the transition from sauropsids into mammals likely occurred only once, with bats being a possible bridge to other mammals (Galen et al., 2018; Lutz et al., 2016). In light of the results of this study and the patterns in other systems, we contend that the ‘bat-seeding’ hypothesis may apply more widely among mammalian parasites. Our approach using comprehensive phylogenetic analysis, estimation of divergence times, and ancestral reconstruction of host associations could be applied to understand the evolutionary radiation and host-switching patterns of these parasites, and potentially the role that bats have played in their diversification.