INTRODUCTION
Since the proposal of DNA barcoding by Hebert et al. (Reference Hebert, Cywinska, Ball and DeWaard2003) as a new methodology for identification of biological species, it has been utilized on a wide variety of taxa for the purposes of identifying museum specimens, evaluation of population and community diversity, discovery of cryptic species, and other forensic applications. Barcoding was adapted for species-level identification by recovery of short DNA sequences from a specific genome fragment and has been applied widely in processing and identifying animal and plant tissues. Attempts have also been made to apply this barcoding paradigm to other eukaryote taxa and some microscopic organisms (Blaxter, Reference Blaxter2016). For example, the nuclear ribosomal internal transcribed spacer (ITS) region has been recognized as a potentially universal DNA barcode marker for fungi (Bellemain et al. Reference Bellemain, Carlsen, Brochmann, Coissac, Taberlet and Kauserud2010; Schoch et al. Reference Schoch, Seifert, Huhndorf, Robert, Spouge, Levesque, Chen and Consortium2012). Nevertheless, acceptance of DNA barcoding for the identification of organisms is controversial among some taxonomists because of fears that a universal, DNA-based approach in identification of species will replace traditional methods (Lebonah et al. Reference Lebonah, Dileep, Chandrasekhar, Sreevani, Sreedevi and Pramoda Kumari2014).
Although bacterial DNA analysis has been readily accepted for measuring the assembly, diversity and distribution of entire microbial communities in different environments (DeLong and Pace, Reference DeLong and Pace2001), application of universal barcoding approaches to bacteria may face challenges unique to prokaryotes. First, many microbial species are challenging to culture by traditional methods (Schloss and Handelsman, Reference Schloss and Handelsman2004), which will curtail a microbiologist's ability to characterize these species beyond molecular approaches. Second, there are still discussions about the precise definition of bacterial species and the numerous, competing methods used to characterize them (Konstantinidis et al. Reference Konstantinidis, Ramette and Tiedje2006; Fraser et al. Reference Fraser, Hanage and Spratt2007). Finally, the phylogenetic diversity observed in microbes is far more complex than in eukaryotes (Hug et al. Reference Hug, Baker, Anantharaman, Brown, Probst, Castelle, Butterfield, Hernsdorf, Amano, Ise, Suzuki, Dudek, Relman, Finstad, Amundson, Thomas and Banfield2016), with predictions of global bacterial diversity ranging from 107 to 1012 species (Curtis et al. Reference Curtis, Sloan and Scannell2002; Schloss and Handelsman, Reference Schloss and Handelsman2004; Dykhuizen, Reference Dykhuizen2005; Locey and Lennon, Reference Locey and Lennon2016). We argue that although our ability to measure bacterial diversity at a massive scale has increased in recent decades, the identification of bacterial species is complex and may not be amenable to one universal approach.
Meanwhile, microbiologists are faced with the dilemma of processing the large bank of specimens sitting in laboratory freezers and museum collections across the globe, containing uncharacterized bacterial species and potential pathogens. These specimens may include irreplaceable tissue samples from rare or hard to reach animal species, samples from human subjects or enormous collections of arthropod ectoparasites. How should microbiologists approach assessing the diversity of bacteria parasitizing these diverse animals in a consistent manner that will facilitate comparative epidemiological, evolutionary and ecological analyses? For the remainder of this review, we will focus on the diverse genus Bartonella (Alphaproteobacteria: Rhizobiales) to make specific recommendations for the genotyping of these bacteria that could be applicable to a wider array of bacterial taxa. Bartonellae are facultative, fastidious, intracellular bacteria commonly found in many taxonomically diverse mammalian species globally (Kosoy et al. Reference Kosoy, Hayman and Chan2012). Bartonellae are hypothesized to be transmitted (some possibly harboured) by a variety of arthropod vectors, including ticks, mites, lice, fleas, flies and other insects (Billeter et al. Reference Billeter, Levy, Chomel and Breitschwerdt2008; Tsai et al. Reference Tsai, Chang, Chuang and Chomel2011). Given the high prevalence and broad host range of Bartonella species, we expect a wide variety of Bartonella species and genotypes to be present in animal tissues collected during field investigations or archived in museum and laboratory collections. Furthermore, this genus exemplifies many of the challenges to characterizing bacterial species, including the limitations of 16S ribosomal RNA (rRNA) sequencing (Kosoy et al. Reference Kosoy, Hayman and Chan2012), challenges with culturing and frequent homologous recombination among genomic loci (Berglund et al. Reference Berglund, Ellegaard, Granberg, Xie, Maruyama, Kosoy, Birtles and Andersson2010; Chaloner et al. Reference Chaloner, Ventosilla and Birtles2011; Paziewska et al. Reference Paziewska, Harris, Zwolińska, Bajer and Siński2011, Reference Paziewska, Siński and Harris2012; Guy et al. Reference Guy, Nystedt, Sun, Näslund, Berglund and Andersson2012; Buffet et al. Reference Buffet, Pisanu, Brisse, Roussel, Félix, Halos, Chapuis and Vayssier-Taussat2013a; Bai et al. Reference Bai, Hayman, McKee and Kosoy2015a). Our analyses in this review will be structured as follows: (1) highlight the value of accurate bacterial genotyping for epidemiological and ecological research, (2) address the challenge of identifying a bacterial species in animal tissues using a short sequence of one or a few selected genomic fragments, (3) analyse the literature on molecular identification of bartonellae in different animal tissues and from diverse animal taxa, (4) compare genetic markers used for genotyping bartonellae with and without culturing and (5) evaluate the phylogenetic resolution of candidate loci based on analysis of genomes available for multiple Bartonella species. In the Discussion, we will make specific recommendations for consistent methods for genotyping bartonellae that will facilitate comparative studies.
NEED FOR ACCURATE GENOTYPING OF BARTONELLA SPECIES IN ANIMAL TISSUES
Importance of genotyping bartonellae in epidemiology and for biological threat preparation
Genotyping of pathogenic zoonotic bacteria can be a very important part of epidemiological investigations in an effort to define the source of human and animal diseases. The importance of tracing pathogenic bacteria in environment was highlighted by threats presented by select biological agents, particularly anthrax (Keim et al. Reference Keim, Pearson and Okinaka2008). In a more common situation, extensive databases containing sequences of multiple bacterial strains can provide irreplaceable information for a comparison of gene sequences between a presumptive human pathogen and potential zoonotic sources. As an example related to the study of bartonellae, a recent case of lymphadenopathy in Tbilisi, Georgia was linked to infected rats (Kandelaki et al. Reference Kandelaki, Malania, Bai, Chakvetadze, Katsitadze, Imnadze, Nelson, Harrus and Kosoy2016). This was only possible because of the accurate genetic characterization of related Bartonella strains from commensal rats in Israel by Harrus et al. (Reference Harrus, Bar-Gal, Golan, Elazari-Volcani, Kosoy, Morick, Avidor and Baneth2009). Similar connections have been made between human cases of myocarditis and meningitis in the USA and Bartonella genotypes found in ground squirrels (Kosoy et al. Reference Kosoy, Murray, Gilmore, Bai and Gage2003; Osikowicz et al. Reference Osikowicz, Billeter, Rizzo, Rood, Freeman, Burns, Hu, Juieng, Loparev and Kosoy2016). Genotyping bartonellae and other infectious bacteria using consistent molecular approaches that generate a common repertoire of gene sequences will surely increase the feasibility and frequency of these comparisons.
Bartonellae as a popular tool for ecological studies
Bartonella species span the symbiont–pathogen continuum (Segers et al. Reference Segers, Kešnerová, Kosoy and Engel2017) and are an extremely diverse group of bacteria, especially in rodents and bats (Lei and Olival, Reference Lei and Olival2014). Moreover, these vertebrate host–arthropod vector–Bartonella systems appear to be globally distributed and phylogenetically complex. Such features make these tripartite systems a popular tool for ecological comparative analyses (Buffet et al. Reference Buffet, Kosoy and Vayssier-Taussat2013b; Klangthong et al. Reference Klangthong, Promsthaporn, Leepitakrat, Schuster, McCardle, Kosoy and Takhampunya2015; Brook et al. Reference Brook, Bai, Yu, Ranaivoson, Shin, Dobson, Metcalf, Kosoy and Dittmar2017). There are some ecological projects where identification of Bartonella is not essential and where a priority is given to estimation of Bartonella prevalence in animal populations without identification of the species (Bai et al. Reference Bai, Kosoy, Calisher, Cully and Collinge2009; Young et al. Reference Young, Dirzo, Helgen, McCauley, Billeter, Kosoy, Osikowicz, Salkeld, Young and Dittmar2014). However, most ecological and epidemiological studies require accurate identification of specific bacterial species and/or genotypes. The level of discrimination between obtained strains or genotypes depends on the objectives of the studies. In most situations, the investigators prefer to report bacteria at the species level or compare sequence identity with a specific Bartonella type strain. There is however a potential pitfall in reporting PCR-positive samples without sequencing of positive products. Some ecologists interested in using simple techniques for estimating prevalence of common animal infection may not be aware that the primers and real-time PCR probes selected for molecular detection of Bartonella DNA may not always be specific for the genus Bartonella (Maggi and Breitschwerdt, Reference Maggi and Breitschwerdt2005). In the absence of sequence data, reporting PCR-positive samples alone may overestimate Bartonella prevalence in such ecological studies. Therefore, we advocate that studies of Bartonella prevalence, and ideally all surveys of infectious bacteria, should adhere to the standard of reporting only sequence-positive samples. Besides the clarity of the results, assessing the diversity and prevalence of Bartonella species in ecological studies with sequence-based approaches will facilitate comparisons of bacterial prevalence and diversity across various spatial and temporal scales and among host species and communities. These analyses are necessary for a deeper understanding of the ecology and evolution of Bartonella species and other infectious bacteria.
CURRENT APPROACHES TO GENOTYPING BARTONELLAE
Modern microbiologists rely heavily on molecular techniques and associated thresholds to assess the diversity of bacteria in various environments and classify new species. Sequencing of the 16S rRNA gene (Woese and Fox, Reference Woese and Fox1977) and DNA–DNA hybridization experiments (Wayne et al. Reference Wayne, Brenner, Colwell, Grimont, Kandler, Krichevsky, Moore, Moore, Murray, Stackebrandt, Starr and Trüper1987) are two methods capable of delineating bacterial species that were developed fairly early, yet in the intervening years, the limitations of these approach have become more clear (Konstantinidis et al. Reference Konstantinidis, Ramette and Tiedje2006). We will highlight some of these limitations below, and where applicable, make specific connections to the study of bartonellae.
Is bacterial isolation a necessary step for genotyping of bartonellae?
The isolation of Bartonella bacteria from infected animals is the preferred method for the diagnosis and characterization of species (Gutiérrez et al. Reference Gutiérrez, Vayssier-Taussat, Buffet and Harrus2017). Gutiérrez et al. (Reference Gutiérrez, Vayssier-Taussat, Buffet and Harrus2017) discuss various methods that are successful in culturing bartonellae and make recommendations for particular sample types (tissue vs ectoparasite). However, one persistent challenge is that many microbes will be challenging to isolate with known culturing techniques. We do acknowledge that cultured isolates are crucially important to the identification of bacterial species and obtaining a culture will facilitate all varieties of morphological, biochemical and genetic analyses. There are projects already underway to sequence the whole genome of all known bacterial-type strains (Kyrpides et al. Reference Kyrpides, Hugenholtz, Eisen, Woyke, Göker, Parker, Amann, Beck, Chain, Chun, Colwell, Danchin, Dawyndt, Dedeurwaerdere, DeLong, Detter, De Vos, Donohue, Dong, Ehrlich, Fraser, Gibbs, Gilbert, Gilna, Glöckner, Jansson, Keasling, Knight, Labeda and Lapidus2014), and these data will no doubt expand our knowledge of bacterial diversity and genomic architecture. For bartonellae specifically, culturing can be very time-consuming due to the slow growth of the bacteria (which can be complicated by overgrowth of other contaminating bacteria) and may not be able to detect some species that do not grow quickly on standard blood agar. Thus, a strictly culture-based assessment may be severely biased towards cultivable strains.
The 16S ribosomal RNA sequencing and the problem of identification of Bartonella species
Assessing bacterial diversity using 16S rRNA sequences has become a very popular technique, especially with the advent of high-throughput sequencing instruments (e.g. Roche 454, Illumina MiSeq/HiSeq and Ion Torrent). This approach has been able to uncover an enormous diversity of bacterial and archaeal taxa, some of it consisting of heretofore uncultured microbial ‘dark matter’ (Rinke et al. Reference Rinke, Schwientek, Sczyrba, Ivanova, Anderson, Cheng, Darling, Malfatti, Swan, Gies, Dodsworth, Hedlund, Tsiamis, Sievert, Liu, Eisen, Hallam, Kyrpides, Stepanauskas, Rubin, Hugenholtz and Woyke2013; Saw et al. Reference Saw, Spang, Zaremba-Niedzwiedzka, Juzokaite, Dodsworth, Murugapiran, Colman, Takacs-Vesbach, Hedlund, Guy and Ettema2015), in environments ranging from mammalian guts and feces (Manichanh et al. Reference Manichanh, Rigottier-Gois, Bonnaud, Gloux, Pelletier, Frangeul, Nalin, Jarrin, Chardon, Marteau, Roca and Dore2006; Bittar et al. Reference Bittar, Keita, Lagier, Peeters, Delaporte and Raoult2014) and parasitic arthropods (Qiu et al. Reference Qiu, Nakao, Ohnuma, Kawamori and Sugimoto2014; Razzauti et al. Reference Razzauti, Galan, Bernard, Maman, Klopp, Charbonnel, Vayssier-Taussat, Eloit and Cosson2015) to marine habitats (Logares et al. Reference Logares, Sunagawa, Salazar, Cornejo-Castillo, Ferrera, Sarmento, Hingamp, Ogata, de Vargas, Lima-Mendez, Raes, Poulain, Jaillon, Wincker, Kandels-Lewis, Karsenti, Bork and Acinas2014).
Nevertheless, there is substantial evidence that 16S rRNA sequencing (and its high-throughput applications) may be inadequate for accurately identifying bacterial species and may not be sensitive enough to recover the complex evolutionary histories of many microbial species. Assessment of microbial taxonomic diversity using 16S rRNA sequences commonly follows a threshold of 97% sequence identity to delineate operational taxonomic units, which can obscure the distinction between closely related species and even genera (e.g. Escherichia and Shigella). Many authors have determined that Bartonella species exhibit very high levels of 16S rRNA gene sequence similarity (Birtles and Raoult, Reference Birtles and Raoult1996). Comparing sequences of 17 Bartonella species and subspecies, La Scola et al. (Reference La Scola, Zeaiter, Khamis and Raoult2003) reported the lowest discriminatory power (99·7%) and highest interspecies similarity (99·8%) for 16S rRNA, making this genetic locus an ineffective tool for the systematic classification of related bacterial species.
Metagenomics of microbial communities and needs for Bartonella genotyping
While reporting the low discriminatory power of 16S rRNA for identification of Bartonella species, La Scola et al. (Reference La Scola, Zeaiter, Khamis and Raoult2003) acknowledged that this locus is still reliable for differentiation of all Bartonella species from Brucella species (94% similarity), the genus taxonomically closest to Bartonella. This fact can justify the application of ribosomal primers for identification of bartonellae as components of microbial communities using 16S rRNA amplicon sequencing. Few surveys based on metagenomic evaluation of rodent-associated bacteria, including bartonellae, were conducted recently (Razzauti et al. Reference Razzauti, Galan, Bernard, Maman, Klopp, Charbonnel, Vayssier-Taussat, Eloit and Cosson2015; Galan et al. Reference Galan, Razzauti, Bard, Bernard, Brouat, Charbonnel, Dehne-Garcia, Loiseau, Tatard, Tamisier, Vayssier-Taussat, Vignes and Cosson2016; Koskela et al. Reference Koskela, Kalin-Mänttäri, Hemmilä, Smura, Kinnunen, Niemimaa, Henttonen and Nikkari2017). These studies highlight the utility of metagenomic techniques; however, they are not without their own limitations regarding the distinction among related bacterial species and potential sequence amplification biases.
Razzauti et al. (Reference Razzauti, Galan, Bernard, Maman, Klopp, Charbonnel, Vayssier-Taussat, Eloit and Cosson2015) compared two next-generation sequencing approaches (transcriptome RNA sequencing and 16S metagenomics) according to their ability to survey multiple bacteria in rodent populations in the French Ardennes region. Among vector-borne bacteria, Bartonella was the most prevalent (>5 reads in 89% of the rodents by 16S sequencing). The authors acknowledged that an important limitation of these approaches is the accuracy of the taxonomic assignation. RNA sequencing allowed taxonomic classification at the species level, while 16S metagenomics classification was generally restricted to the genus level. Analysing this problem, Razzauti et al. (Reference Razzauti, Galan, Bernard, Maman, Klopp, Charbonnel, Vayssier-Taussat, Eloit and Cosson2015) stressed the point that the 16S rRNA gene is difficult to sequence in its totality because of the size (~1550 bp) using current high-throughput sequencing methods. The method proposed by Miller et al. (Reference Miller, Baker, Thomas, Singer and Banfield2011) allows to assembly steps, but is not frequently used because of the increased experimental complexity and cost. Instead, a portion of the 16S rRNA gene is usually amplified using specific sets of universal primers. The nine hypervariable (V) regions of the 16S rRNA gene differ between species, and depending on the V region chosen, one can discriminate some species but not others. In their paper, Razzauti et al. (Reference Razzauti, Galan, Bernard, Maman, Klopp, Charbonnel, Vayssier-Taussat, Eloit and Cosson2015) have also reported an interesting observation about a large difference in the relative abundance of Bartonella reads detected by the 16S MiSeq (95%) vs RNA sequencing (<1%).
Galan et al. (Reference Galan, Razzauti, Bard, Bernard, Brouat, Charbonnel, Dehne-Garcia, Loiseau, Tatard, Tamisier, Vayssier-Taussat, Vignes and Cosson2016) investigated the potential for recent developments in 16S rRNA-based high-throughput sequencing (Illumina MiSeq) to facilitate the multiplexing of urban rodents in West Africa. This study reported significant difference in Bartonella prevalence between rodent species varying from 0·5% in Mus musculus to 79% in Mastomys natalensis. Praising advances in this screening strategy, the authors admit that 16S rRNA amplicon sequencing based on a short sequence did not yield results sufficiently high in resolution to distinguish between Bartonella species (Galan et al. Reference Galan, Razzauti, Bard, Bernard, Brouat, Charbonnel, Dehne-Garcia, Loiseau, Tatard, Tamisier, Vayssier-Taussat, Vignes and Cosson2016). Another metagenomic evaluation of bacteria in voles from Finland (Koskela et al. Reference Koskela, Kalin-Mänttäri, Hemmilä, Smura, Kinnunen, Niemimaa, Henttonen and Nikkari2017) reported commonality of Bartonella species in the voles, although identification of the species was not clear. André et al. (Reference André, Mouton, Millien and Michaux2017) used 16S metagenomics to investigate the liver microbiome of Peromyscus leucopus mice in Canada, finding no difference between the microbiome assemblages of mouse genotypes separated by the Saint Lawrence River. In contrast to the other studies above, the authors used an additional marker (16S–23S intergenic spacer, ITS) to identify all of the Bartonella species as B. vinsonii arupensis, a known zoonotic agent in humans (Welch et al. Reference Welch, Carroll, Hofmeister, Persing, Robison, Steigerwalt and Brenner1999; Bai et al. Reference Bai, Kosoy, Diaz, Winchell, Baggett, Maloney, Boonmar, Bhengsri, Sawatwong and Peruski2012).
Banskar et al. (Reference Banskar, Bhute, Suryavanshi, Punekar and Shouche2016) used 16S metagenomic sequencing (Ion Torrent) to investigate the fecal microbiome of Rousettus leschenaultii bats in India. They found a high abundance of Proteobacteria in some of the samples, which contains a large number of pathogenic genera, including Bartonella. However, the authors claim to have detected Bartonella henselae in two of the bat samples, which is highly unlikely given the strong association of B. henselae with cats. This misidentification is most likely due to the authors’ use of a 97% sequence identity threshold, which is insufficient to distinguish among Bartonella species. Dietrich et al. (Reference Dietrich, Kearney, Seamark and Markotter2017) also applied 16S metagenomic sequencing (Illumina MiSeq) to characterize the microbiome in saliva, urine and feces from four species of insectivorous bats from South Africa. Similarly to Banskar et al. the authors found a high abundance of Proteobacteria in bat feces, but also in saliva and urine. Sequences mapping to Bartonella were found predominantly in feces, but also to some extant in saliva and urine. No attempts were made to identify the specific Bartonella species found in these samples; however, as we have discussed above, this would likely not be possible using only 16S sequences.
In review, the utility of 16S sequencing will largely depend on the questions investigators wish to pursue, and the scale of phylogenetic resolution needed to answer such questions. If investigators wish to assess bacterial diversity in specimens at the genus level or higher, 16S metagenomics would be an excellent approach. As we have reviewed above, below the genus level however, this gene will not be sufficient to accurately distinguish among related species. Depending on the focus of the study, investigators could then target a few genera of interest for characterization with more discriminating genetic loci (André et al. Reference André, Mouton, Millien and Michaux2017). For example, investigators may target genera with high abundance in the 16S dataset that may contain pathogenic species. This approach has been used with success recently to describe Bartonella species in bats (Veikkolainen et al. Reference Veikkolainen, Vesterinen, Lilley and Pulliainen2014; Wilkinson et al. Reference Wilkinson, Duron, Cordonin, Gomard, Ramasindrazana, Mavingui, Goodman and Tortosa2016).
DNA–DNA hybridization and its limitations for identification of bacteria
The DNA–DNA hybridization experiments used by Wayne et al. (Reference Wayne, Brenner, Colwell, Grimont, Kandler, Krichevsky, Moore, Moore, Murray, Stackebrandt, Starr and Trüper1987) represented a potentially more robust approach for delineating bacterial species using the whole genome. A threshold of 70% hybridization has been used as the ‘gold standard’ criterion for distinguishing new bacterial species (Tindall et al. Reference Tindall, Rosselló-Móra, Busse, Ludwig and Kämpfer2010); however, this technique requires the use of cultured isolates, specialized equipment and multiple confirmatory tests due to variation across experimental runs. Alternative genome-wide distance measures included average nucleotide identity (Konstantinidis and Tiedje, Reference Konstantinidis and Tiedje2005; Konstantinidis et al. Reference Konstantinidis, Ramette and Tiedje2006; Goris et al. Reference Goris, Konstantinidis, Klappenbach, Coenye, Vandamme and Tiedje2007; Richter and Rosselló-Móra, Reference Richter and Rosselló-Móra2009) and digital DNA–DNA hybridization (Auch et al. Reference Auch, von Jan, Klenk and Göker2010; Meier-Kolthoff et al. Reference Meier-Kolthoff, Hahnke, Petersen, Scheuner, Michael, Fiebig, Rohde, Rohde, Fartmann, Goodwin, Chertkov, Reddy, Pati, Ivanova, Markowitz, Kyrpides, Woyke, Göker and Klenk2014a, Reference Meier-Kolthoff, Klenk and Gökerb). One considerable advantage of these techniques is that they do not necessarily require cultured isolates and can be calculated from draft genomes assembled from metagenome and transcriptome sequencing. However, as noted above, these high-throughput techniques are currently not accessible to very many research groups and will not have much utility unless investigators have a bacterial isolate or a draft genome.
Recombination as an important complication for genotyping
Another important issue that can arise when attempting to genotype a bacterial strain is that separate genes may indicate the presence of different species. These conflicts arise due to lateral gene transfer (LGT) among bacteria, either directly through conjugation or indirectly via phage-mediated transduction or transformation by uptake of free DNA in the environment. LGT is the predominant mechanism by which bacteria acquire antibiotic resistance genes and can be an important part of bacterial evolution (Vos, Reference Vos2009). Homologous recombination is a specific form of LGT whereby homologous genes of a donor genome replace the gene variant in the recipient genome. Homologous recombination is a common feature among some bacterial species (Vos and Didelot, Reference Vos and Didelot2008) and even among distantly related bacteria (Hanage et al. Reference Hanage, Spratt, Turner and Fraser2006), thus severely complicating phylogenetic inference. This problem has been documented in several studies of Bartonella strains from cats, rodents and bats based on sequencing multiple protein-coding loci (Berglund et al. Reference Berglund, Ellegaard, Granberg, Xie, Maruyama, Kosoy, Birtles and Andersson2010; Chaloner et al. Reference Chaloner, Ventosilla and Birtles2011; Paziewska et al. Reference Paziewska, Harris, Zwolińska, Bajer and Siński2011, Reference Paziewska, Siński and Harris2012; Guy et al. Reference Guy, Nystedt, Sun, Näslund, Berglund and Andersson2012; Buffet et al. Reference Buffet, Pisanu, Brisse, Roussel, Félix, Halos, Chapuis and Vayssier-Taussat2013a; Bai et al. Reference Bai, Hayman, McKee and Kosoy2015a) and have provided valuable information about the mechanisms that generate Bartonella diversity and the gene flow among co-occurring species. We note here that these studies have been limited to cultured strains. As we will discuss later in the paper, attempts to genotype bartonellae from genomic DNA extracted from whole blood, tissue or from ectoparasites may be further complicated by the presence of multiple Bartonella species in the sample. However, sequencing multiple loci will clarify if recombination or multiple species are present and phylogenetic concordance among sequenced loci can be sufficient to describe a potentially novel Bartonella species or subspecies.
Gene-sequence-based paradigm for identification of Bartonella isolates
There are existing methods, particularly multi-locus sequence typing (MLST; Stackebrandt et al. Reference Stackebrandt, Frederiksen, Garrity, Grimont, Kämpfer, Maidem, Nesme, Rosselló-Morá, Swings, Trüper, Vauterin, Ward and Whitman2002), which can balance the tradeoffs of culturing bias, phylogenetic resolution, homologous recombination and gene conservation across species. MLST of house-keeping genes (i.e. genes under stabilizing selection encoding metabolic functions) remains a powerful technique that can be used on uncultured bacteria to detect evidence of mixed infections and/or homologous recombination, provide sufficient phylogenetic resolution for the delineation of bacterial species, and will provide consistency in the usage of genetic loci that can facilitate global assessments of parasitic bacterial diversity.
Using such an approach, La Scola et al. (Reference La Scola, Zeaiter, Khamis and Raoult2003) compared the similarities of seven genetic loci among the 17 species and subspecies of genus Bartonella. This comparison led to both the definition of similarity values that discriminated Bartonella at the species level and assessment of the relative discriminatory power of each gene examined. The gltA, groEL, rpoB and ftsZ genes, and ITS all have good discriminating power ranging from 92·6 to 94·4%. Overall, two genes (rpoB and gltA) were found to be the most potent markers for demarcation of Bartonella species (La Scola et al. Reference La Scola, Zeaiter, Khamis and Raoult2003). This paper was very influential for characterization of Bartonella cultures and defining their status as a species. Many studies have now used MLST approaches to characterize Bartonella genotypes and species, and based on the available methods, some form of multi-locus sequencing appears to be the most viable method available to most researchers.
Confirmation of species status based on comparison of gene profiles among related species
Characterizing the novel Bartonella species (B. melophagi) isolated from sheep blood and sheep keds, Kosoy et al. (Reference Kosoy, Bai, Enscore, Rizzo, Bender, Popov, Albayrak, Fofanov and Chomel2016) reported presence of 183 genes specific for this species, being absent in genomes of other Bartonella species associated with ruminants to support their argument for the separation of this bacterial species from species of other ruminant-associated Bartonella species. The authors identified that out of the 1338 genes, the number of homologous but unique genes was estimated to be 1274, out of which 156 genes appeared to be specific to B. melophagi and absent in any of the 21 reference genomes Bartonella species. Comparison of the gene profile of B. melophagi with related Bartonella species associated with ruminants (B. bovis and B. schoenbuchensis) demonstrated that 183 genes were present only in the genome of B. melophagi, while 1027 genes present in one or more copies in each genome were conserved between these three bacterial strains. The remaining 27 genes were present in B. melophagi and absent in related species (B. bovis and B. schoenbuchensis), but found in at least one of the other Bartonella species. This analysis indicates that even among related bacterial species, genomes can be very flexible in gene content, and can be useful criterion for describing novel species (Konstantinidis and Tiedje, Reference Konstantinidis and Tiedje2005; Konstantinidis et al. Reference Konstantinidis, Ramette and Tiedje2006).
Applying multi-locus approaches to detection and sequencing without cultured isolates
Due to the challenges of culturing bartonellae, investigators may choose to characterize Bartonella infections in animal samples directly from extracted DNA. Despite the convenience of such an approach, it has its own challenges. For instance, some primers used to amplify genetic loci may be insufficiently sensitive to amplify Bartonella DNA from some animal tissues. The presence of PCR inhibitors that carry through from blood, tissue or the extraction process may also interfere with detection. The use of nested PCR reactions may be able to overcome some of these deficiencies among sensitivity among loci for detection and may not require much additional primer design. For instance, protocols exist for amplifying gltA and ftsZ sequences using nested reactions with known primer sets (Norman et al. Reference Norman, Regnery, Jameson, Greene and Krause1995; Birtles and Raoult, Reference Birtles and Raoult1996; Zeaiter et al. Reference Zeaiter, Liang and Raoult2002; Colborn et al. Reference Colborn, Kosoy, Motin, Telepnev, Valbuena, Myint, Fofanov, Putonti, Feng and Peruski2010; Gundi et al. Reference Gundi, Billeter, Rood and Kosoy2012a).
Another challenge with amplification directly from extracted DNA is the potential presence of multiple bacterial species in the sample, and coinfections of multiple Bartonella genotypes in one animal are not uncommon based on culturing. When multiple species are present in a sample, the abundance of their DNA may vary in the sample and even across tissue types. Furthermore, different primer sets may have amplification bias towards particular species based on the annealing affinity. These complications may cause the observed Bartonella diversity to differ depending on which marker was used for amplification, so a single marker may not be a robust indicator of total Bartonella diversity in a set of samples (Buffet et al. Reference Buffet, Pisanu, Brisse, Roussel, Félix, Halos, Chapuis and Vayssier-Taussat2013a). Furthermore, investigators have observed recombination events even within a single gene (gltA), interfering with phylogenetic inference, but may not be present in other sequenced loci (Paziewska et al. Reference Paziewska, Siński and Harris2012; Buffet et al. Reference Buffet, Pisanu, Brisse, Roussel, Félix, Halos, Chapuis and Vayssier-Taussat2013a).
Nevertheless, this approach has one very important disadvantage – when sequenced loci are in conflict regarding the bacterial species identified in the sample, one must determine if this is caused by homologous recombination or the presence of multiple infections. In these cases, researchers may choose to report the conflicting results as is and simply note this caveat with the understanding that culturing, cloning sequences into vectors before sequencing, or deep sequencing approaches may differentiate these possible scenarios. In some cases, multiple peaks may be visible in chromatograms of sequences, so cloning would be useful in these cases, but not all cases of multiple infections show this pattern, probably due to varying abundances of DNA that are not detected in the consensus sequence reads.
Even with these known limitations, this multi-locus sequencing approach has been used recently to detect and characterize Bartonella genotypes from bats, rodents and carnivores. Lilley et al. (Reference Lilley, Veikkolainen and Pulliainen2015) used a combination of rpoB and gltA sequences to characterize a novel Bartonella species (Candidatus B. hemsundetiensis) in Myotis daubentonii bats from Finland. Similar Bartonella species have subsequently been cultured and characterized from related insectivorous bats in the Republic of Georgia (Urushadze et al. Reference Urushadze, Bai, Osikowicz, McKee, Kandaurov, Kuzmin, Sidamonidze, Putkaradze, Imnadze and Kosoy2017). Martin-Alonso et al. (Reference Martin-Alonso, Houemenou, Abreu-Yanes, Valladares, Feliu and Foronda2016) used multiple loci (ITS, gltA and rpoB) to detect Bartonella infections in rodent species from Benin. Based on these markers, the authors describe a distinct Bartonella species (Candidatus B. mastomydis) from M. natalensis. Sequences very similar to this candidate species had previously been acquired from related rodent species in Ethiopia. The authors also reported the presence of multiple peaks in their sequencing results, so they used cloning to distinguish the coinfections. However, there were additional conflicts between gltA and rpoB sequences for some samples that did not show multiple sequence peaks, with one locus indicating the presence of B artonella elizabethae and the other indicating B artonella tribocorum (Martin-Alonso et al. Reference Martin-Alonso, Houemenou, Abreu-Yanes, Valladares, Feliu and Foronda2016). The authors hypothesize that these conflicts may have arisen by recombination; however, as we noted above, multiple infections (with no evidence of multiple sequence peaks) may be an alternative explanation.
These studies, although fairly recent, demonstrate the potential of this multi-locus approach to characterizing bartonellae without a culturing step. New Bartonella species can be described across multiple genes showing phylogenetic concordance, or in other cases, interesting cases of potential recombination or multiple infection can be noted. In this way, multi-locus sequencing can be an important first step, focusing on detection and partial genotyping, with other analyses following after to fully characterize novel or recombinant genotypes by culturing and MLST (or full genome analyses). Multi-locus sequencing therefore strikes a valuable balance by providing potentially more robust assessments of Bartonella diversity than single-locus approaches, and is also more accessible to a wider community of researchers than full genomic approaches since it requires only standard molecular techniques (PCR and Sanger sequencing).
Metagenome and transcriptome sequencing – the way of the future?
Shotgun metagenome and transcriptome sequencing techniques that target many coding loci are becoming popular methods for identifying bacteria at the species level with better phylogenetic resolution at low per-base cost (Venter et al. Reference Venter, Remington, Heidelberg, Halpern, Rusch, Eisen, Wu, Paulsen, Nelson, Nelson, Fouts, Levy, Knap, Lomas, Nealson, White, Peterson, Hoffman, Parsons, Baden-Tillson, Pfannkoch, Rogers and Smith2004; Rinke et al. Reference Rinke, Schwientek, Sczyrba, Ivanova, Anderson, Cheng, Darling, Malfatti, Swan, Gies, Dodsworth, Hedlund, Tsiamis, Sievert, Liu, Eisen, Hallam, Kyrpides, Stepanauskas, Rubin, Hugenholtz and Woyke2013; Logares et al. Reference Logares, Sunagawa, Salazar, Cornejo-Castillo, Ferrera, Sarmento, Hingamp, Ogata, de Vargas, Lima-Mendez, Raes, Poulain, Jaillon, Wincker, Kandels-Lewis, Karsenti, Bork and Acinas2014; Hug et al. Reference Hug, Baker, Anantharaman, Brown, Probst, Castelle, Butterfield, Hernsdorf, Amano, Ise, Suzuki, Dudek, Relman, Finstad, Amundson, Thomas and Banfield2016) and may represent a new way forward for identifying bacterial pathogens in a large number of samples and tissue types. However, the cost of deep sequencing and absence of comprehensive reference databases (the majority of environmental microorganisms have yet to be sequenced) making this approach unavailable for all but the most well-funded laboratories. We look forward to seeing these high-throughput metagenome and transcriptome approaches applied more frequently (and we expect they will as the cost of machinery and computing resources become more available and affordable), but for now we seek to make recommendations for genotyping bartonellae that are accessible to a wider community of researchers. As noted above, a multi-locus sequence approach may be the best option for many studies and could facilitate broad-scale comparisons of Bartonella diversity across systems if researchers use a consistent set of markers.
REVIEW OF STUDIES FOCUSING ON GENOTYPING BARTONELLAE FROM ANIMALS SAMPLES
In order to make recommendations for sequence-based approaches for genotyping bartonellae from archived animal samples, we performed a literature review to identify commonly used genetic markers. Based on the results of this survey, we will identify some candidate markers that could become consistent features of the multi-locus sequencing approach we described above, and thus facilitate valuable comparative studies of Bartonella ecology and evolution.
Analysis of literature on identification of bartonellae in animal hosts
We surveyed a sample of published literature (>400 studies) using paired key words ‘bartonella-rodents’, ‘bartonella-bats’, ‘bartonella-wildlife’, ‘bartonella-cats’, ‘bartonella-dogs’ and ‘bartonella-ectoparasites’. Of the processed literature, 293 studies were selected with available information on application of diverse genetic markers for genotyping of bartonellae in identified tissues of vertebrate animals and/or their ectoparasites. These studies report investigations conducted in 79 countries of Africa (19), the USA (11), Asia (21), Australia/Oceania (5) and Europe (23), and from a broad diversity of animal taxa, including rodents, bats, carnivores, ruminants and marine mammals (Table 1). In the 101 studies, bartonellae were cultured from blood, followed by genotyping of the isolates. Of those, culturing work was accompanied with molecular detection of Bartonella DNA in tissues only in 16 studies, while detection of Bartonella DNA in ectoparasites along with culturing bacteria from their hosts was attempted in 21 studies.
Table 1. Studies aimed to identifying bartonellae in different tissues of vertebrate animals and their ectoparasites (only studies with identified animal hosts are selected; studies with testing off-hosts arthropods are excluded)
a Cattle, sheep, goats, horses, camels.
b Deer, moose, feral pigs.
c Coyotes, foxes, jackals, raccoons, others.
d Seals, dolphins, porpoises, sea otters.
e Exotic animals, marsupials, primates, birds, turtles.
Selection of animal tissues for detection and genotyping of Bartonella species
The most frequent tissue for targeting and genotyping Bartonella DNA by PCR and sequencing was blood: 62 studies where only blood was used and three studies where other tissues along with blood were analysed (Table 1). Other tissues used for Bartonella genotyping are: spleen (29), liver (8), heart (7), kidney (5), lung (2), and ear, skin and nail by one study. Besides blood samples, only one tissue type was analysed in 34 studies, two tissues in eight studies and more than two tissues in three studies.
Overall, there are limited reports about significant variation in detection of Bartonella DNA between tissues. Razzauti et al. (Reference Razzauti, Galan, Bernard, Maman, Klopp, Charbonnel, Vayssier-Taussat, Eloit and Cosson2015) noted that the choice of organ likely has an important impact on the detection or misdetection of Bartonella. To explain the huge difference in relative abundance of Bartonella reads detected by 16S MiSeq vs RNA-Seq cited above, the authors used the currently accepted model of Bartonella infection described by Harms and Dehio (Reference Harms and Dehio2012). This model posits that immediately after infection, bartonellae colonize an unknown primary niche in the mammalian host, most likely vascular endothelial cells. Every 5 days, some of the bacteria in the endothelial cells are released into the blood stream, where they infect erythrocytes. Then bacteria invade a phagosomal membrane inside the erythrocytes, where they multiply until they reach a critical population density. At this point, they simply wait until they are taken up with the erythrocytes by a blood-sucking arthropod. The spleen plays important roles with regard to erythrocytes by removing old erythrocytes, and may thereby hold a reserve of erythrocytes that are highly infected by non-replicating bartonellae, which do not produce RNA molecules. Moreover, due to its central role in recycling erythrocytes, the spleen could also store a large amount of degraded DNA of dead bartonellae (Razzauti et al. Reference Razzauti, Galan, Bernard, Maman, Klopp, Charbonnel, Vayssier-Taussat, Eloit and Cosson2015).
Genetic markers used for identification of bartonellae
Based on our review, the total number of genetic loci that have been used for genotyping Bartonella DNA, either from bacterial culturing or tissue extracts, reached 41 (Fig. 1a). The applied markers include both coding genes and intergenic regions. From 1994 when the genotyping of Bartonella was initiated through 2002, the procedure of genotyping was limited to the application of two markers (gltA and 16S). The number of genetic loci used has steadily increased after 2002 (Fig. 1a), although the majority of studies only use one or two markers (Fig. 1c). Most of the genetic loci were used in only a few studies and were not repeated in laboratories other than ones where they were proposed. Only 10 genetic loci were used >10 times in multiple laboratories (gltA, ITS, rpoB, 16S rRNA, ftsZ, groEl, ribC, pap31, nuoG and ssrA) with gltA being the most frequently used marker across all studies surveyed (Fig. 1a).
Fig. 1. Frequency of genetic loci used for genotyping bartonellae (a), temporal trend in the number of markers used for genotyping (b) and the frequency distribution of the number of markers used across 293 reviewed studies.
Beyond detection, genetic targets that provide sufficient sequence diversity to allow differentiation of Bartonella species are required to fully understand the distribution and host specificity of various Bartonella species and identification of the strains associated with human illness. The citrate synthase gene (gltA), originally proposed by Norman et al. (Reference Norman, Regnery, Jameson, Greene and Krause1995), remains the most popular genetic target for Bartonella detection and is considered a reliable tool for distinguishing genotypes. In our review, gltA was used in 48 of 56 of the studies where one or two markers were applied for identification of Bartonella cultures [Table 2(a)]. Other markers, particularly rpoB and ftsZ, are common when at least four markers are used for direct detection of Bartonella DNA in tissues by PCR; ITS is used frequently, comparable with the gltA and more often than rpoB and ftsZ [Table 2(b)].
Table 2. Summary of markers used for the characterization of bartonellae (a) from cultures, (b) tissues and (c) ectoparasites
Few attempts have been made to culture bartonellae from arthropod ectoparasites, so identification of Bartonella from ectoparasites is typically performed by PCR on extracted DNA. Studies have identified Bartonella DNA in a number of ectoparasite groups: fleas (80), ticks (40), lice (13), bat flies (9), deer and sheep keds (9), mites (6), Cimex spp. bugs (2), bees (2) and ants (1). Detection and genotyping primarily target ITS and gltA, with rpoB being the third most common marker [Table 2(c)].
Comparison of genetic markers for detection and genotyping of Bartonella DNA in animal tissues
Of 54 publications where identification and genotyping of bartonellae in mammalian tissues were conducted with at least two different genetic markers, only 13 studies provided data for comparing the effectiveness of using different genetic loci (Table 3). In almost all of these studies, the ITS target was the most sensitive marker for identification of Bartonella DNA in blood. Only one study focused on detecting and genotyping Bartonella DNA in cat blood found the ITS and gltA targets to be equally productive (Bai et al. Reference Bai, Rizzo, Alvarez, Moran, Peruski and Kosoy2015b). While detecting and genotyping Bartonella in rodent spleens, two studies reported successful identification in more specimens by targeting the rpoB gene compared with the gltA (Gundi et al. Reference Gundi, Kosoy, Myint, Shrestha, Shrestha, Pavlin and Gibbons2010, Reference Gundi, Kosoy, Makundi and Laudisoit2012b).
Table 3. Comparison of genetic loci used for genotyping of Bartonella DNA in animal tissues by their detection frequency
a The genetic markers used as the second step for genotyping Bartonella in positive DNA after initial screening.
b Real-time PCR assay was used for detection of Bartonella DNA without genotyping.
c Data are not in publication, but provided from a private communication.
Birtles and coworkers described the use of PCR-based amplification of ITS fragments to detect and identify bartonellae in the blood of rodents. Direct detection was of particular use in the longitudinal survey of Bartonella bacteraemia that involved the field collection of very small amounts of blood from live, wild rodents (Birtles et al. Reference Birtles, Hazel, Bown, Raoult, Begon and Bennett2000). As most of the ITS is non-coding, it is prone to hypervariability, and its sequence variation is markedly higher than that observed at other genetic loci (Roux and Raoult, Reference Roux and Raoult1995). Although comparison of ITS sequences is useful for the allocation of detected organisms into one of the recognized Bartonella species, detection of a novel ITS sequence can be problematic because of the difficulties with sequencing of amplified fragments and problems with alignment of the obtained sequences (Knap et al. Reference Knap, Duh, Birtles, Trilar, Petrovec and Avšič-Županc2007). ITS sequences have many insertions and deletions that can complicate phylogenetic analysis. In some of the studies, screening was conducted by conventional or real-time PCR of ITS, followed by sequencing of additional markers, usually gltA (Miceli et al. Reference Miceli, Gavioli, Goncalves, Andre, Sousa, de Sousa and Machado2013; Gutiérrez et al. Reference Gutiérrez, Nachum-Biala and Harrus2015; Bai et al. Reference Bai, Gilbert, Fox, Osikowicz and Kosoy2016).
Comparison of genetic markers for detection and genotyping of Bartonella DNA in ectoparasites
Since only few attempts of culturing Bartonella from arthropods have been successful, genotyping of Bartonella in insects and acarines relies mostly on detection from extracted DNA by PCR, typically relying on only one marker. In spite of a large number of publications reporting investigation of Bartonella in ectoparasites (>140), we were able to select only 14 publications, which provided data on comparison of at least two genetic markers for genotyping Bartonella in DNA extracted from arthropods (Table 4). In four of the 14 studies, ITS was shown to be most sensitive marker for detection and genotyping Bartonella; however, in some other studies, success with gltA gene was similar (De Sousa et al. Reference De Sousa, Edouard-Fournier, Santos-Silva, Amaro, Bacellar and Raoult2006; Pérez-Martínez et al. Reference Pérez-Martínez, Venzal, González-Acuña, Portillo, Blanco and Oteo2009).
Table 4. Comparison of genetic loci used for genotyping of Bartonella DNA in animal ectoparasites by their detection frequency
a Data are not in publication but provided from a private communication.
b Real-time PCR assay was used for detection of Bartonella DNA without genotyping.
When Morick et al. (Reference Morick, Krasnov, Khokhlova, Shenbrot, Kosoy and Harrus2010) genotyped bartonellae in fleas collected from rodents in the Negev Desert of Israel using three genetic markers (gltA, ITS and rpoB), they found the 313 bp gltA fragment to be the best target for screening fleas for Bartonella and for identification to species level. All flea pools that were found positive by rpoB or ITS screening were also positive by gltA. Pérez-Martínez et al. (Reference Pérez-Martínez, Venzal, González-Acuña, Portillo, Blanco and Oteo2009) investigated 82 fleas collected from cats and dogs in Chile. When rpoB primers were used, Bartonella genotypes were found in four Ctenocephalides felis fleas from cats (4·8%) and in four Pulex irritans fleas from dogs (4·8%). The same eight samples were positive when primers for gltA and ITS were used. None of the 82 specimens were positive when primers targeting the groEL gene were used. Conducting surveillance of Egyptian fleas for agents of public health significance, Loftis et al. (Reference Loftis, Reeves, Szumlas, Abbassy, Helmy, Moriarity and Dasch2006) detected more Bartonella-positive fleas using groEL than ITS (17 vs 11) and were successful in conducting phylogenetic analysis based on comparison of the groEL sequences rather than ITS sequences.
Contribution of analyses of complete Bartonella genomes to primer design
Cross-referencing the gltA primer set against the GenBank dataset showed that despite their common use for Bartonella detection, these primers have high cross-reactivity both to potential Bartonella host DNA (such as Rattus, Mus and Homo sapiens) and to bacterial species that could inhabit similar ecological niches (such as Ehrlichia) (Colborn et al. Reference Colborn, Kosoy, Motin, Telepnev, Valbuena, Myint, Fofanov, Putonti, Feng and Peruski2010). To identify genus-specific and host-blind primer sets, a whole-genome scan of three Bartonella genomes (B. henselae, B. quintana and B. bacilliformis) available at that time was performed (Colborn et al. Reference Colborn, Kosoy, Motin, Telepnev, Valbuena, Myint, Fofanov, Putonti, Feng and Peruski2010), and the NADH dehydrogenase γ subunit (nuoG) primer set was identified and met all the required conditions. A few years later, another genetic locus (ssrA), also known as transfer-messenger RNA, was proposed as a target for a genus-specific real-time PCR assay based on analyses on whole genomes (Diaz et al. Reference Diaz, Bai, Malania, Winchell and Kosoy2012). These markers have been used in a number of studies for the detection of Bartonella DNA in animal tissues and ectoparasites, with successful detection at rates similar to other loci but still lower than ITS (Gutiérrez et al. Reference Gutiérrez, Cohen, Morick, Mumcuoglu, Harrus and Gottlieb2014; Brook et al. Reference Brook, Bai, Dobson, Osikowicz, Ranaivoson, Zhu, Kosoy and Dittmar2015; Bai et al. Reference Bai, Rizzo, Alvarez, Moran, Peruski and Kosoy2015b).
Recommendations for marker usage in a multi-locus sequencing framework
Overall, the Bartonella gltA sequence database in GenBank is the largest and most frequently updated among the different collections of deposited sequences, and therefore allows a more accurate differentiation between Bartonella species and strains. A proteomic analysis of gltA indicates that most amino substitutions are synonymous, highlighting the important and critical function of the citrate synthase (gltA) enzyme. Nevertheless, numerous studies have indicated that ITS is a highly sensitive marker that is invaluable for the detection of Bartonella DNA. In order to maximize detection success and differentiation among related Bartonella species, we advocate for a multi-locus sequencing approach. Although many markers have been used in different studies, there is a growing consensus of frequently used markers – specifically, gltA, ITS, rpoB, ftsZ, ribC, groEL, nuoG and ssrA – that are generally capable of differentiating among Bartonella species, particularly when used together in a multi-locus genotyping framework (La Scola et al. Reference La Scola, Zeaiter, Khamis and Raoult2003). Usage of these markers consistently across studies will facilitate ecological analyses of Bartonella prevalence and diversity across systems and comprehensive phylogenies of known Bartonella species.
EVALUATION OF PHYLOGENETIC RESOLUTION AMONG CANDIDATE LOCI BASED ON ANALYSIS OF BARTONELLA GENOMES
La Scola et al. (Reference La Scola, Zeaiter, Khamis and Raoult2003) used seven protein-coding loci to genotype Bartonella strains, but as we noted above, these genes varied considerably in their power to discriminate among Bartonella species. In the intervening years, genomes of many Bartonella species have been sequenced and assembled. Using these data, we will evaluate the phylogenetic resolution of a number of candidate loci found in the genomes of Bartonella species and compare these results to the genetic markers frequently used to genotype bartonellae.
Detection of gene clusters in Bartonella genomes
Genomes from 22 publically available Bartonella species were downloaded from GenBank. Every pair of gene sequences from each genome was aligned using the Needleman–Wunsch global alignment algorithm to all other genes. The resulting alignment scores were placed in a square similarity matrix and genes were assigned to clusters using a single-linkage (non-centroid-based, non-greedy), exhaustive clustering algorithm. The clustering threshold was chosen in a way so that each gene cluster is expected to contain the same genes originating from different species. The constituent sequences of each gene cluster were then partitioned into separate FASTA files for subsequent analyses (L. Albayrak and C. McKee, unpublished data).
Ranking of gene clusters by sequence diversity
The overall sequence diversity of each gene cluster was estimated as the ratio between the numbers of unique 32-base long subsequences present in all sequences in the cluster over the total number of 32-base long subsequences present in all sequences in the cluster. For each gene cluster, all 32-base long subsequences (32-mers) from each position in the nucleotide sequences in the FASTA file were collected. The reverse complements of the extracted subsequences were added to the complete set of 32-mers. Unique 32-base long subsequences in the set were identified and their ratio to the total number of subsequences (including duplicates) was calculated. We refer to this measurement as the proportion of unique 32-mers, and it varies between near zero and one. This measure is equal to one if all 32-base long subsequences identified in the clusters are unique.
Gene clusters were sorted in descending order based on the proportion of unique 32-mers and assigned numerical ranks accordingly. The top 10 (most diverse) and bottom 10 (most conserved) clusters containing sequences from each of 22 Bartonella genomes were selected. We also identified seven genes commonly used to genotype bartonellae (ftsZ, gltA, groEL, gyrB, nuoG, ribC and rpoB) from the ranking, corresponding to the following ranks out of 665 total gene clusters: 171 (ribC), 386 (gyrB), 543 (nuoG), 565 (gltA), 611 (ftsZ), 635 (rpoB) and 658 (groEL). This resulted in a list of 26 gene clusters since groEL was part of the bottom 10. We then added 16S rRNA (rank 665/665) and ITS (rank 652/665) to this ranking separately for each Bartonella species for which these sequences were available (L. Albayrak and C. McKee, unpublished data). The proportion of unique 32-mers is used here to assess the diversity of the nucleotide sequences in each cluster and is a useful measure for ranking many gene clusters by their sequence diversity. However, this measure does not necessarily reflect phylogenetic differentiation among congeneric taxa. Hence, we then quantified the sequence diversity and phylogenetic resolution of each of these 28 gene clusters using additional measures. Sequence diversity was assessed based on the proportion of segregating sites, Watterson's estimator of genetic diversity and nucleotide diversity. Phylogenetic resolution was measured by calculating Tamura-Nei sequence distances and storing the minimum, median and maximum distances. All calculations for these measures were performed in MEGA (Kumar et al. Reference Kumar, Stecher and Tamura2016).
Across all of the 28 gene clusters, other measures of sequence diversity generally followed a declining trend that corresponded to the ranking by proportion of unique 32-mers (Fig. 2a), and all of the measures were moderately to highly correlated (0·65 < r < 1). However, there was some variation present in these estimates that was not captured in the proportion of unique 32-mers, particularly in the proportion of segregating sites. Tamura-Nei sequence distances similarly declined across the ranking of gene clusters, with some noticeable variation in the median and maximum distances (Fig. 2b). The nine genetic loci we analysed (16S rRNA, ITS, ftsZ, gltA, groEL, gyrB, nuoG, ribC and rpoB) fell between the top 10 and bottom 10 based on the proportion of unique 32-mers, with the exception of groEL and 16S, which had the eighth lowest and the lowest rankings, respectively. These two regions also had low sequence diversity by other measures and low Tamura-Nei distances, indicating that they have poor phylogenetic resolution. Overall, the top 10 candidate loci do show significantly higher measures of sequence diversity and phylogenetic distance than the 18 other loci (Fig. 3); however, the distributions of minimum Tamura-Nei distance among these groups do overlap (Fig. 1e). These minimum distances correspond to the inverse of the maximum sequence similarity that La Scola et al. (Reference La Scola, Zeaiter, Khamis and Raoult2003) used to assess discriminatory power among loci. Generally, these minimum distances are small among all loci, ranging from 0·002 for 16S rRNA to just 0·085 for the top-ranked gene cluster, an unnamed membrane protein (Fig. 2b). Among the eight other commonly used markers (ITS, ftsZ, gltA, groEL, gyrB, nuoG, ribC and rpoB), the minimum distances ranged were 0·012 for groEL to 0·038 for gyrB and 0·045 for ITS. The majority of these minimum distances were between B. melophagi and B. schoenbuchensis, two Bartonella species found in ruminants (deer and sheep).
Fig. 2. Measures of sequence diversity (a) and phylogenetic distance (b) across 28 genetic loci occurring in all 22 Bartonella genomes currently available. Loci are numerically ranked in descending order along the bottom axis according to the proportion of unique 32-mers. Commonly used markers for genotyping are labelled with their gene names next to their numerical rank.
Fig. 3. Comparisons of the proportion of unique 32-mers (a), proportion of segregating sites (b), Watterson's estimator (c), nucleotide diversity (d), Tamura-Nei sequence distance (minimum, median and maximum); (e–f) across 28 genetic loci occurring in all 22 Bartonella genomes currently available. Separate boxes in each plot represent the top 10 loci as ranked by the proportion of unique 32-mers (‘top 10’) and the other 18 loci (‘remaining’).
Our results largely confirm what La Scola et al. (Reference La Scola, Zeaiter, Khamis and Raoult2003) found; however, our rankings of the markers with the ability to distinguish closely related species were somewhat different, and this is partly due to our usage of entire gene sequences for our measurements (La Scola et al. used only partial gene sequences). In both of our analyses, 16S rRNA displays the lowest ability to discriminate among Bartonella species. The other eight genetic loci commonly used for genotyping perform much better than 16S rRNA, with minimum distances exceeding 1%, which should be sufficient for identifying two genotypes as distinct in a phylogenetic analysis, particularly when used together in MLST analyses.
We were able to identify the presence of candidate loci that can discriminate among closely related Bartonella species better than the commonly used markers. We suggest that these loci may be useful for the characterization of Bartonella species and assessment of phylogenetic relationships; however, there currently exist no known primers for amplifying these loci by conventional PCR. Bartonella species have high nucleotide diversity across their genomes (Fig. 2a) with few highly conserved regions, especially in highly diverse genes. Therefore, primer design is a very challenging problem, especially the design of universal primers capable of binding to all possible species (L. Albayrak and C. McKee, unpublished data). There is likely a tradeoff in between phylogenetic resolution of genes and the ability to design universal primers, so the clustering of the eight genetic loci commonly used for genotyping (ITS, ftsZ, gltA, groEL, gyrB, nuoG, ribC and rpoB) between the most diverse genes and the least diverse genes (including 16S rRNA) may be a function of this tradeoff. The other potential disadvantage of using any of the top 10 most diverse loci for genotyping bartonellae is that these genes have only been sequenced for 22 Bartonella species. There are an enormous number of potentially new Bartonella species and genotypes that have been characterized by only ITS and/or gltA sequences that considerably expand our knowledge of Bartonella diversity; thus, there is an advantage to continued usage of these markers to facilitate comparative ecological analyses. Switching to different markers would inevitably ignore this diversity and would require considerable time and effort to restore.
DISCUSSION
The practicality of single-locus barcoding of bacteria
The utility of DNA barcoding for animal species is partly due to special features of the genetic markers it targets. The commonly targeted cytochrome c oxidase I (COI) gene for barcoding animals is one of the conserved oxidative phosphorylation subunits of the mitochondrial genome. Mitochondria are passed solely from the female parent to offspring in animals, so individuals are typically haploid at all mitochondrial loci and no recombination occurs in the mitochondrial genome. Additionally, there are sufficiently conserved portions of the COI gene that nearly universal PCR primers have been developed (Hajibabaei et al. Reference Hajibabaei, Janzen, Burns, Hallwachs and Hebert2006). Since bacterial genomes are haploid, it may be tempting to simply find any genetic marker that has better phylogenetic resolution than 16S rRNA and use it for DNA barcoding of bacteria. There are four primary problems with this approach: (1) increasing sequence diversity in a gene diminishes the ability to design conserved primers that can be utilized across a broad taxonomic diversity of bacteria; (2) homologous recombination is widespread (Vos and Didelot, Reference Vos and Didelot2008) and can obscure phylogenetic inference (Fraser et al. Reference Fraser, Hanage and Spratt2007) and estimates of species diversity if only one marker is used; (3) bacterial genomes are very flexible in gene content, even for closely related species (Konstantinidis et al. Reference Konstantinidis, Ramette and Tiedje2006), so a locus may not exist in all species surveyed; and (4) multiple species may be present in the sample, but may not be detectable due to variation in abundance or primer amplification bias. Thus, we argue that for bacterial species, and in particular Bartonella, there is probably no perfect analogue to single-locus DNA barcoding that could be used for all sample types.
There are existing methods, particularly MLST (Stackebrandt et al. Reference Stackebrandt, Frederiksen, Garrity, Grimont, Kämpfer, Maidem, Nesme, Rosselló-Morá, Swings, Trüper, Vauterin, Ward and Whitman2002), which could balance the tradeoffs of culturing bias, phylogenetic resolution, homologous recombination and gene conservation across species. We believe that MLST of house-keeping genes (i.e. genes under stabilizing selection encoding metabolic functions) remains a powerful technique that can be used on uncultured bacteria to detect evidence of mixed infections and/or homologous recombination, provide sufficient phylogenetic resolution for the delineation of bacterial species and will provide consistency in the usage of genetic loci that can facilitate a global assessments of parasitic bacterial diversity. This approach can be appropriately modified for the detection and characterization of parasitic bacteria directly from extracted DNA from a range of sample types, with the caveat that culturing should be attempted if feasible since it is the best way to fully characterize a novel bacterial species. In the following section, we will make specific recommendations for the detection and genotyping Bartonella in collected samples (Box 1), but we recognize that with some modifications to collection of appropriate samples and the molecular protocols, this approach is likely generalizable to a variety of bacterial taxa.
Box 1. Recommendations for consistent approaches to genotyping bartonellae
(a) Target multiple animal tissues (e.g. blood, spleen, liver and/or heart) and ectoparasites for detection. Identify vertebrate and arthropod hosts to the species level where possible using morphological traits and/or barcoding of mtDNA to facilitate ecological analyses.
(b) Use homogenization and DNA extraction protocols that maximize yield while reducing the presence of PCR inhibitors. Extended lysis or pre-enrichment culture steps may be needed for some samples. After homogenization of samples but before DNA extraction, retain some samples if culturing will be attempted.
(c) When possible, attempt to culture isolates, especially when sequence data indicate the presence of novel species or genotypes. Genotyping should be regarded as just the first step towards the description of Bartonella species, with additional trait and genomic data providing valuable information for species descriptions.
(d) Screen samples by ITS or real-time PCR (ITS, rpoB or ssrA) or alternatively, 16S metagenome or transcriptome sequencing (if available) followed by sequencing of multiple house-keeping genes.
(e) At the very least, sequence gltA to facilitate comparison with other studies.
(f) Sequence at least one additional marker to confirm the species identity based on gltA. Conflicting identifications may indicate the presence of multiple infections or recombinant genotypes. More markers provide more robust results, but at least three is recommended.
(g) Additional targets can vary in detection success, but rpoB, ftsZ, groEL, ribC, nuoG and ssrA are popular (in order of frequency used). Nested PCR reactions can increase sensitivity of these markers to be more comparable with ITS results.
(h) Attempt to identify the phylogenetic lineage (Harms and Dehio, Reference Harms and Dehio2012) or associated Bartonella species complex (Kosoy et al. Reference Kosoy, Hayman and Chan2012) based on sequence data for any novel genotypes.
Recommendations for the genotyping of Bartonella species
The first step to successful detection and potential isolation of bartonellae is to collect appropriate tissues and store them properly [Box 1(a)]. Gutiérrez et al. (Reference Gutiérrez, Vayssier-Taussat, Buffet and Harrus2017) recommend the collection of whole blood due to the haemotrophic nature of these bacteria using appropriate sterility requirements, especially if culturing is to be attempted. If animals are sacrificed and organs pulled out, spleen is probably the most valuable organ for Bartonella detection, although as we have reviewed above, liver, heart, kidney and lung may show evidence of infection, and Bartonella species may vary in abundance across these tissue types within individuals. All tissue samples should be transported at low temperature and stored at −20 or −80 °C if not processed immediately. For ectoparasite samples, storage in 70% ethanol at room temperature is convenient and suitable for molecular detection; however, if culturing is planned, then live specimens are preferred and additional surface sterilization protocols will be required (Gutiérrez et al. Reference Gutiérrez, Vayssier-Taussat, Buffet and Harrus2017). During the process of collecting and analysing specimens, we recommend that investigators attempt to identify all animals as close as possible to the species level by morphological traits [Box 1(a)]. When morphological identification is not feasible (e.g. when cryptic species of rats and ectoparasites are morphologically undistinguished or when accurate records do not exist), then DNA barcoding of host samples (tissues or whole ectoparasites) at mitochondrial loci (e.g. COI) can be incorporated into molecular analyses (Hajibabaei et al. Reference Hajibabaei, Janzen, Burns, Hallwachs and Hebert2006). These data are valuable for understanding the ecology and evolution of Bartonella species, particularly for understanding the host range and specificity, vector potential and evolutionary codivergence of parasites with their hosts and vectors.
We recommend that investigators use homogenization techniques appropriate for particular tissue or ectoparasite specimens, and follow extraction protocols that maximize DNA yield and quality while minimizing the presence of PCR inhibitors that may be present in the specimens [Box 1(b)]. Some specimens may benefit from pre-enrichment in liquid growth medium before extraction (Maggi et al. Reference Maggi, Duncan and Breitschwerdt2005; Duncan et al. Reference Duncan, Maggi and Breitschwerdt2007; Riess et al. Reference Riess, Dietrich, Schmidt, Kaiser, Schwarz, Schafer and Kempf2008; Bai et al. Reference Bai, Kosoy, Boonmar, Sawatwong, Sangmaneedet and Peruski2010) or extended lysis steps during the extraction process. Gutiérrez et al. (Reference Gutiérrez, Vayssier-Taussat, Buffet and Harrus2017) provide an excellent review of recommended protocols for homogenization and extraction. In cases where culturing might be attempted, we recommend retaining samples of homogenate (either used immediately or frozen at −20 °C or below). As we have advocated above, culturing is vital for the complete description of bacterial species and should be attempted in all studies where appropriate samples are available, especially if sequence data indicate the presence of novel Bartonella species or genotypes [Box 1(c)]. Culturing can provide information on valuable traits such as in vitro growth rate, bacterial morphology, presence of multiple coinfecting bartonellae, biochemical profiles, etc. Additional genomic (e.g. MLST or whole genome sequencing) analyses that clarify evolutionary histories can be facilitated, if Bartonella genotypes are isolated.
For direct detection of Bartonella DNA from extracted DNA, there are several options for markers that appear to have good sensitivity. Conventional PCR targeting the 16S–23S intergenic spacer region (ITS) or real-time PCR targeting various loci (e.g. ITS, ssrA, rpoB) are amenable for screening many samples for potential positives [Box 1(d)]. Primers and protocols for these approaches are published and are reviewed in Gutiérrez et al. (Reference Gutiérrez, Vayssier-Taussat, Buffet and Harrus2017). We caution against reporting results from real-time PCR assays or conventional PCR in the absence of sequencing, since not all primers are entirely specific for Bartonella DNA (Maggi and Breitschwerdt, Reference Maggi and Breitschwerdt2005; Colborn et al. Reference Colborn, Kosoy, Motin, Telepnev, Valbuena, Myint, Fofanov, Putonti, Feng and Peruski2010; Diaz et al. Reference Diaz, Bai, Malania, Winchell and Kosoy2012) and may amplify host DNA, leading to false positives. Usage of next-generation sequencing approaches (e.g. 16S metagenomics, transcriptomics) are also valuable at this stage, especially if investigators are interested in describing the microbiome of particular tissues or investigating a broad range of pathogenic bacterial taxa [Box 1(d)]. However, as we reviewed above, the phylogenetic resolution of 16S sequences is limited, so additional genomic loci will need to be sequenced to confirm the species identity of targeted bacterial taxa.
For accurate genotyping, there are a variety of markers that have good phylogenetic resolution (La Scola et al. Reference La Scola, Zeaiter, Khamis and Raoult2003) and validated primer sets (Gutiérrez et al. Reference Gutiérrez, Vayssier-Taussat, Buffet and Harrus2017). Based on our review of the literature, gltA is the most widely used marker and has the most extensive database of sequences available on GenBank. Therefore, we advocate for all studies to sequence this marker, at least from any novel genotypes or species, to support comparisons of Bartonella diversity across studies [Box 1(e)]. Nested PCR reactions (Bai et al. Reference Bai, Gilbert, Fox, Osikowicz and Kosoy2016) that combine published primers (Norman et al. Reference Norman, Regnery, Jameson, Greene and Krause1995; Birtles and Raoult, Reference Birtles and Raoult1996) can increase the sensitivity of this marker.
In addition to gltA, we feel it is important to acquire additional sequence data from other loci to confirm the species identification by gltA [Box 1(f)]. Phylogenetic concordance among loci may be sufficient to describe novel, candidate Bartonella species (Lilley et al. Reference Lilley, Veikkolainen and Pulliainen2015; Martin-Alonso et al. Reference Martin-Alonso, Houemenou, Abreu-Yanes, Valladares, Feliu and Foronda2016), which can be further described after culturing isolates. Conflicts among multiple loci may indicate the presence of multiple infections or recombinant infections – molecular cloning or examination of multiple peaks in chromatograms may be able to distinguish these scenarios. In general, three loci should be sufficient to accurately genotype Bartonella if a single infection is present or detect mixed infections; however, more loci may produce more robust results or distinguish Bartonella genotypes that may be closely related to known species. The number of markers to use will depend on the phylogenetic resolution needed (most MLST studies require only 5–9 markers), and more markers, especially uncommon markers, may have limited usefulness if not repeated in other laboratories. Additional targets can vary in sensitivity and amplification bias for particular species [Box 1(g)], but rpoB, ftsZ, groEL, ribC, nuoG and ssrA are popular (in order of frequency used based on our literature review) and have published primers and amplification protocols (Gutiérrez et al. Reference Gutiérrez, Vayssier-Taussat, Buffet and Harrus2017), including some nested protocols (Zeaiter et al. Reference Zeaiter, Liang and Raoult2002; Colborn et al. Reference Colborn, Kosoy, Motin, Telepnev, Valbuena, Myint, Fofanov, Putonti, Feng and Peruski2010; Bai et al. Reference Bai, Gilbert, Fox, Osikowicz and Kosoy2016). Our analysis of 22 Bartonella genomes indicates that there may be single loci that are capable of distinguishing among Bartonella species better than these popular markers alone, but primer design for these regions will be challenging due to their sequence diversity (i.e. with few conserved regions), and the utility of these sequences will lag behind these popular markers unless many laboratories adopt them. Researchers should experiment with multiple markers and modified protocols to find the best ones for their sample type, but some consistency among sequenced markers across studies will facilitate comparative studies of Bartonella prevalence and diversity across systems.
Finally, our understanding of Bartonella ecology and evolution would benefit from the increased description of Bartonella phylogenetic lineages (Harms and Dehio, Reference Harms and Dehio2012) and species complexes (Kosoy et al. Reference Kosoy, Hayman and Chan2012), especially for novel genotypes [Box 1(h)]. Species complexes can include clusters genetically similar species found in a group of related hosts, such as B. elizabethae, Bartonella queenslandensis, Bartonella rattimassiliensis and B. tribocorum associated with murine rodents. These species complexes can indicate the presence of evolutionary radiations through codivergence and speciation within related hosts, providing information about the biological niche of these Bartonella species. Multiple species complexes linked deeper in evolutionary time may form well-supported clades or lineages that illuminate the longer term diversification processes of this diverse genus. Researchers describing new Bartonella species or genotypes could increase the impact of their findings by making these substantial evolutionary and ecological connections, particularly when clinical cases of bartonellosis can be traced to a potential zoonotic origin.
Concluding remarks
Using Bartonella bacteria as examples, we have highlighted the substantial challenges that exist in the accurate genotyping of bacteria from environmental samples. Issues related to isolation of cultures, homologous recombination, coinfections, sensitivity and phylogenetic resolution of molecular markers, variation in detection across different tissues, and variation in marker usage across studies are certainly not restricted to studies of Bartonella. For environmental samples stored in laboratories and museums around the world, our recommendations for sensitive detection assays (including real-time PCR or high-throughput metagenomics), followed by conventional PCR amplification and sequencing of multiple house-keeping genes are surely applicable to a wide array of other zoonotic bacteria. These may include many proteobacteria (e.g. Anaplasma, Bordatella, Brucella, Burkholderia, Campylobacter, Coxiella, Ehrlichia, Yersinia, Francisella, Helicobacter, Legionella, Neorickettsia, Orientia, Pasteurella, Pseudomonas, Rickettsia and Wolbachia), spirochetes (e.g. Borrelia, Leptospira, Treponema) and other bacteria (e.g. Chlamydia, Listeria, Mycobacterium, Mycoplasma). Although recommendations for the collection of appropriate animal samples and molecular markers used for detection and characterization will vary according to each bacteria (and likely requires standardization as we observed with Bartonella), the general process of direct detection from extracted DNA, multi-locus sequencing and subsequent attempts to culture (followed by additional genomic or biochemical characterization) is generalizable. Databases already exist, e.g. http://www.mlst.net/databases/default.asp and https://pubmlst.org/databases/), that contain primers and protocols for multi-locus sequencing approaches for many of the zoonotic bacteria listed above. Standardized genotyping approaches will greatly expand our knowledge of the phylogenetic diversity and ecology of parasitic bacteria infecting animals, and help to measure and mitigate the risks posed by these bacteria to public health.
ACKNOWLEDGMENTS
The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of CDC.
FINANCIAL SUPPORT
Numerous investigations of zoonotic bacterial agents around the globe were supported by the CDC’ Global Disease Detection program.
Open access
