Borrelia burgdorferi, the agent of Lyme disease, disseminates from the site of deposition by Ixodes ticks to cause systemic infection. Dissemination occurs through the circulation and through tissue matrices, but the B. burgdorferi molecules that mediate interactions with the endothelium in vivo have not yet been identified. In vivo selection of filamentous phage expressing B. burgdorferi protein fragments on the phage surface identified several new candidate adhesins, and verified the activity of one adhesin that had been previously characterized in vitro. P66, a B. burgdorferi ligand for β3-chain integrins, OspC, a protein that is essential for the establishment of infection in mammals, and Vls, a protein that undergoes antigenic variation in the mammal, were all selected for binding to the murine endothelium in vivo. Additional B. burgdorferi proteins for which no functions have been identified, including all four members of the OspF family and BmpD, were identified as candidate adhesins. The use of in vivo phage display is one approach to the identification of adhesins in pathogenic bacteria that are not easily grown in the laboratory, or for which genetic manipulations are not straightforward.
Phage display is a technology that allows the identification of a gene or random nucleotide sequence of interest through the selection of particular properties of the protein or peptide encoded. The DNA encoding the peptide sequence selected is packaged within the same bacteriophage particle, allowing rapid isolation of the portion of the gene encoding the activity of interest. This approach has proven to be a powerful tool for the identification of peptides with particular binding characteristics, and of recombinant antibodies that recognize a particular antigen of interest (examples include McCafferty et al., 1990; Clackson et al., 1991; Lowman et al., 1991; Koivunen et al., 1994; Healy et al., 1995; Coburn et al., 1999). The technology has also allowed the identification of small peptide sequences that bind to the vascular endothelium in particular tissues in a living animal. This ‘in vivo’ phage display approach has demonstrated that the endothelium in different tissues bears distinctive surface markers, and consequently, can be targeted specifically (Pasqualini and Ruoslahti, 1996; Arap et al., 1998; Rajotte et al., 1998). The identification of a peptide that bound to the endothelium in tumour tissue allowed the conjugation of a synthetic form of that peptide to the chemotherapeutic agent doxorubicin, resulting in the delivery of the toxic drug specifically to the tumour (Arap et al., 1998). These studies therefore lay the groundwork for the application of phage display technology to analyses of protein–receptor interactions in other fields, including infectious diseases.
Borrelia burgdorferi is the causative agent of Lyme disease, currently the most common arthropod-borne infection in the United States and some other countries in the northern hemisphere. A key feature of B. burgdorferi is its ability to cause persistent infection in multiple tissues, which is essential to its maintenance in the tick-rodent cycle in nature, but has unfortunate consequences for accidental hosts outside this cycle, including humans. The ability to cause disseminated infection is facilitated by the ability of B. burgdorferi to invade, and then migrate out of, the vascular system. In early, disseminated infection, B. burgdorferi can be found in blood, but is never abundant (Barbour and Hayes, 1986; Wormser et al., 2005). The organism is thought to be cleared from the blood by the host immune response and by the organism's propensity for invasion of vascular walls and colonization of perivascular connective tissues. In infected laboratory mice, the bacteria are most commonly found in or around arterial vessel walls (Barthold et al., 1991; 1993).
Although tremendous strides have recently been made in the genetic manipulation of, and generation of mutants in, B. burgdorferi (reviewed in Cabello et al., 2001; Elias et al., 2003), there are still a number of challenges to the identification of genes important to the virulence of this organism using genetic approaches such as signature-tagged mutagenesis or in vivo expression technology. The transformation of infectious B. burgdorferi is hindered by a relatively low efficiency due to the presence of two putative restriction/modification systems encoded by the loci bbe02 and bbq67 (Kawabata et al., 2004), by the segmented genome composed of a linear chromosome plus numerous linear and circular plasmids (Fraser et al., 1997; Casjens et al., 2000), by the frequent loss of plasmids required for infectivity during in vitro culture, and by the requirement that each transformant clone to be tested for infectivity first be screened to ensure that the full complement of plasmids is still present (Purser and Norris, 2000; Elias et al., 2002; Lawrenz et al., 2002; Grimm et al., 2003). Furthermore, the efficiency of recovery of B. burgdorferi from infected animal tissues appears to be low, and the infection persists in immunocompetent animals with relatively low numbers of organisms present in infected tissues as compared with organisms that, for example, colonize and cause disease in the intestinal tract.
Phage display was previously used in vitro to identify the B. burgdorferi protein, P66, that binds β3-chain integrins (Coburn et al., 1999). Other approaches have been used to identify B. burgdorferi proteins that bind to fibronectin (Probert and Johnson, 1998), glycosaminoglycans (Parveen and Leong, 2000) and decorin (Guo et al., 1995). However, the specific role of P66 during animal infection remains unknown, and Bgp (the Borreliaglycosaminoglycan binding protein) is not required for mammalian infection (Parveen et al., 2006). Two recent reports tested the role of the fibronectin binding protein, BBK32, in the life cycle of B. burgdorferi. One study found that, at a single relatively high infectious dose, there was no significant difference between the wild-type and knockout strains in murine infection, or in bacterial acquisition by or transmission by feeding ticks (Li et al., 2006). In contrast, the other study found that at lower doses, the mutants were attenuated in mice (Seshu et al., 2006). The role of binding to decorin has been explored using decorin-deficient mice, in which arthritis severity, and survival of the bacteria in the skin and joint, are diminished (Brown et al., 2001; Liang et al., 2004a). Recent work suggested that the decorin binding proteins (DbpA and DbpB) are not required for infectivity in mice (Shi et al., 2006), but one single dose of bacteria was used to infect the mice, so whether the dbp- strains might be attenuated could not be determined, and in fact, the dbp- mutants were recovered from mouse tissues less efficiently compared with the wild type. The fibronectin binding protein, BBK32, also binds to glycosaminoglycans (Fischer et al., 2006), and so may functionally overlap with the DbpA, DbpB and Bgp proteins. B. burgdorferi also binds to type I collagen in the native lattice form (Zambrano et al., 2004), but the protein responsible is not yet known. Because B. burgdorferi expresses several different adhesion activities that functionally overlap in vitro, it may be difficult to establish the pathogenic role of any one particular adhesin during infection. In addition, cell lines in culture do not necessarily reflect the properties of the cells from which they derive in the physiologically relevant setting of a living animal. To determine whether any of the known adhesins might be functional in vivo, and to identify additional candidate adhesins that might mediate these interactions but are not expressed by laboratory-cultivated B. burgdorferi, we selected for B. burgdorferi proteins that bind to the endo-thelium in three different tissues using in vivo phage display.
Results and discussion
The subset of B. burgdorferi proteins (and genes) selected in vivo
To identify B. burgdorferi proteins that mediate interactions of the bacteria with the endothelium in different tissues during infection, filamentous bacteriophage clones encoding B. burgdorferi proteins fused to the phage protein III were selected in three tissues that are known to be colonized by the organism. Although many other tissues are infected by B. burgdorferi, limitation to this subset facilitated the prompt processing required to recover phage particles. A schematic representation of the protocol is shown in Fig. 1. Not all of the phage pools that had been through selection demonstrated the highest titres in the tissues in which they were selected (i.e. for a tibiotarsus-selected pool, titre was not highest in the tibiotarsus); those that did not were not further analysed. The latter set is nevertheless likely to contain tissue-selected clones, as in most cases, the titres were highest in the spleen, followed by the tissue of selection. A total of nine heart-selected pools (six of which were randomly chosen for further analysis), three tibiotarsus-selected (hereafter referred to as joint-selected) pools and one bladder-selected phage pool were obtained. These pools demonstrated enrichment of bacteriophage particles from the library on the basis of phage titre relative to titres in the other tissues analysed. It is unclear why the different tissues appeared to have different selective potentials, e.g. why we did not obtain nine tibiotarsus-selected pools with the highest titres in the tibiotarsus, as was the case with the heart. The nine heart-selected phage pools were derived from all three unselected library pools, the three joint-selected pools derived from two of the three input pools, and the bladder-selected pool represented one starting pool. All of the tissue-selected phage pools contained single clones encoding fragments of many B. burgdorferi proteins, representing the background ‘noise’ of this system. For a protein (and its corresponding gene) to be considered to be selected above the background, it had to have been enriched from multiple input library pools, or represented by multiple overlapping clones derived from the same pool. Only a small subset of the B. burgdorferi proteome (and genome) was represented among the selected clones that fit these criteria.
Of the 842 predicted genes (769 of > 300 bp, 73 of ≤ 300 bp; Casjens et al., 2000) in the B. burgdorferi chromosome, we obtained at least two independent isolates of 43 (5.1%). Of the 665 predicted genes (535 of > 300 bp, 130 of ≤ 300 bp; Casjens et al., 2000) contained in the plasmids of B. burgdorferi strain B31 clone M1, we obtained at least two independent phage clones representing 16 (2.4%). This calculation does not include the predicted pseudogenes in the B. burgdorferi strain B31 clone M1 plasmid genome, which are almost entirely plasmid-borne. All pools yielded single isolations of additional genes that were not further analysed.
Proteins known or predicted to be on the B. burgdorferi surface and encoded on the chromosome
The genes encoding proteins that are known to be, or are predicted to be, localized on the surface of B. burgdorferi are of greatest interest in terms of how the bacteria might interact with the mammalian host. The surface proteins selected by in vivo phage display are listed in Table 1. Three genes encoded on the chromosome and eight encoded by plasmids were selected in vivo. The three chromosomal genes, BB0210, BB0385 and BB0603, each have features of interest.
Table 1. Known or predicted outer surface proteins of B. burgdorferi selected by in vivo phage display.
The name appears as listed in the genome site at TIGR, with apologies to the authors who have given some of these proteins other names.
BBK2.10 was originally identified in B. burgdorferi strain 297 as a homologue of OspF of strain N40 (Akins et al., 1995). The clones obtained in our in vivo selection are clearly more related to BBK2.10 than to the originally identified OspF.
The first chromosomally encoded candidate adhesin gene identified, BB0210, encodes a very large protein that is predicted to be localized in the outer membrane. The localization of the protein, however, has not been experimentally demonstrated, and the mature protein is not predicted to contain any transmembrane helices. The function of this protein has not been determined in B. burgdorferi, and it is not clear whether the protein is expressed by the bacteria in the tick host or in laboratory culture, but serum antibodies from B. burgdorferi-infected mice do recognize the selected fragment of BB0210 in recombinant form, suggesting that the protein is expressed by the bacteria during mammalian infection (M. LaFrance and S. Antonara, unpub. data). Of the predicted 1119 amino acids, all clones selected in vivo in the heart, joint and bladder shared a common region of 49 amino acids. This sequence comprises most of the repeat module predicted in the B. burgdorferi B31 M1 sequence, and the B31 protein is predicted to contain seven repeats (Fig. S1).
The second chromosomal locus selected by in vivo phage display, bb0385, encodes a putative lipoprotein termed BmpD. This is one member of a family of four B. burgdorferi proteins that are expressed during mammalian infection, but whose functions are not yet known (Ramamoorthy et al., 1996; Bryksin et al., 2005). The in vivo-selected clones encode the carboxyl-terminal half of BmpD (Fig. S2) plus an extension that appears to be encoded by strain N40 clone D10E9 but not by B31 M1 (the sequenced strain). Localization of BmpD on the surface of B. burgdorferi has not been definitively demonstrated, but the protein is recognized by antibodies from human patients (Bryksin et al., 2005).
BB0603, which was present in heart- and joint-selected pools, encodes P66, which was previously shown to bind to the β3-chain integrins, αIIbβ3 and αvβ3 (Coburn et al., 1999). P66 also binds to α3β1 and weakly to α5β1 (J. Coburn, unpublished, and Coburn et al., 1999). The portion common to the clones selected in vivo overlaps with the portion of the protein previously shown to be required for integrin binding (Fig. 2) (Coburn et al., 1999; Defoe and Coburn, 2001), suggesting that the integrin binding domain is required for binding to the endothelium. Integrin αIIbβ3 is expressed exclusively by platelets and megakaryocytes. Platelets adherent to the endothelium in regions of microvascular damage (e.g. around the heart valves) might therefore serve as sites of attachment for P66, and intact B. burgdorferi cells have been noted in perivascular connective tissue in regions of vascular damage or stresses (Johnston et al., 1985; Barthold et al., 1991; 1993; Armstrong et al., 1992). Integrins αvβ3, α3β1 and α5β1 are expressed by diverse cell types. Although only two clones encoding P66 fragments were selected in vivo, they were selected independently from different starting pools and from different tissues. These observations, together with the previously characterized integrin binding activity of P66, support the hypothesis that binding to integrins might play a role during the course of B. burgdorferi infection and association with vessel walls.
Proteins known or predicted to be on the B. burgdorferi surface and encoded on the plasmids
Several B. burgdorferi plasmid-encoded proteins were also selected by in vivo phage display for attachment to the endothelium in vivo. Two plasmid lp-54-encoded proteins (BBA52 and BBA66) of unknown function were selected (Table 1). Both were present only in heart-selected pools. BBA52 is predicted to be a non-lipidated outer membrane protein that is downregulated in B. burgdorferi in the ‘host-adapted’ dialysis membrane chamber model in comparison with B. burgdorferi grown in vitro (Brooks et al., 2003), so the significance of BBA52 in mammalian infection is dubious. In contrast, BBA66 is predicted to be a lipoprotein of paralogous family 54, and was recently demonstrated to be surface-exposed and expressed during mammalian infection (Brooks et al., 2006; Clifton et al., 2006). Neither BBA52 nor BBA66 has significant homology to proteins outside the Borrelia genus.
Four of the plasmid-encoded genes listed in Table 1, namely BBM38 (ErpK), BBO39 (ErpL), BBK2.10 and BBS41 (OspG), are the members of the OspF family of proteins (paralogous family 164) encoded on circular plasmids. The sequences that we list as BBK2.10 are most related to this protein, which was originally identified in B. burgdorferi strain 297 as an OspF homologue (Akins et al., 1995). In the B. burgdorferi strain B31 M1 sequence, the closest match is to the locus BBR42 (OspF) (Fraser et al., 1997; Casjens et al., 2000). Although some of these proteins have been previously studied and are known by other names, their functions have remained unclear. Our selection of the four OspF family members by in vivo phage display suggests that this family may mediate adhesion to host molecules. The OspE subfamily of the Erp family of proteins, in contrast to the OspF proteins, comprises a group that has been shown to bind complement regulatory factor H. The four proteins that we identified as candidate endothelial adhesins have not been found to bind factor H (Alitalo et al., 2002).
Alignment of the sequences of selected clones representing BBM38 (ErpK), BBO39 (ErpL), BBK2.10 and BBS41 suggests that two distinct regions of these proteins may have adhesin activity (Fig. S3). OspG and BBK2.10 have both regions, while ErpK and ErpL have one each. It is apparent that our clone of N40 contains not only the ospG gene, but also bbK2.10, which has not previously been noted in other N40 clones. It is possible that the original tick isolate designated ‘N40’ in fact represents a population of at least two distinct B. burgdorferi clones.
Two of the plasmid-encoded proteins that were identified in our in vivo phage selection are of considerable interest in the pathogenesis of B. burgdorferi infection. BBF32 encodes the vls locus, which has been demonstrated to encode a single expression site plus 15 silent cassettes (in the sequenced strain B31 M1) that undergo segmental recombination into the expression site ‘variable regions’, generating an enormous number of potential variants (Zhang et al., 1997; Zhang and Norris, 1998a). Recombination occurs during murine infection but not during in vitro cultivation (Zhang and Norris, 1998b), and the plasmid encoding the vls is required for long-term infection but not for the initial colonization of mammals (Labandeira-Rey et al., 2003; Grimm et al., 2004a), suggesting a critical role for the vls locus in persistent infection of mammals. The segment of vls common to all our selected clones (KAIVDAA; Fig. S4) is similar to a portion of invariable region two (IR2: KEIVEAA) and a portion of invariable region four (IR4: SAIVTAA) of the B. burgdorferi strain B31 M1 sequence (Zhang et al., 1997; Eicken et al., 2002). The sequences flanking the sequence common to all the clones differ, which would be expected after passage of the bacteria through an animal. This variation makes precise alignments of the selected Vls sequences from N40 D10E9 to those of B31 M1 difficult. It appears that the selected clones fall into different subgroups based on the alignments (group 1: clones 42D and 971G; group 2: clones 116 A and 196Q), so it is possible that both IR2 and IR4 are represented. The clone of N40 used to generate the phage library had been passaged through a mouse to demonstrate infectivity when it was isolated in the early 1990s (J.M. Leong, pers. comm.). The crystal structure and antibody accessibility studies of VlsE show that the second and fourth invariable regions are at least partially exposed on the surface of the protein and are recognized by the immune system (Liang and Philipp, 1999; Liang et al., 2000; Eicken et al., 2002). These regions do not appear to be accessible to antibodies in in vitro-cultivated bacteria, but it is possible that they are more accessible when the bacteria are in a mammal, a condition in which the abundant OspA protein is not expressed.
BBB19 encodes OspC, which has been demonstrated to be essential for mammalian infection (Grimm et al., 2004b). Recent work demonstrated that OspC is required exclusively early in infection, as when an ospC-null mutant is complemented with ospC on a shuttle vector, that plasmid can be lost during murine infection (Tilly et al., 2006). However, the bacteria remain viable and can be acquired by ticks, in which they survive the molt between blood meals and migrate to the salivary glands, but cannot establish infection in mice. Our results are consistent with the hypothesis that OspC may facilitate the initial colonization of the mammal by serving as an adhesin. OspC-encoding phages were selected most frequently in the heart, and, given that the protein appears to be essential in the early stages of infection, OspC may bind to a receptor, or a modification thereof, that is expressed by cells of the innate immune system and/or the skin, and in the cardiac endothelium. Two groups that determined the crystal structure of OspC noted that the protein forms a dimer and appears to have a binding site for an unknown ligand (Eicken et al., 2001; Kumaran et al., 2001). All of our clones share 75 amino acids that include the segment from loop 2 through α-helix 4 (Fig. 3A) (Eicken et al., 2001; Kumaran et al., 2001), which comprises much of the surface-exposed portion of the protein (Fig. 3B). The helices involved in dimerization were not included in our selected clones.
Proteins not known or predicted to be localized on the surface of B. burgdorferi
A number of B. burgdorferi genes that are not predicted to encode surface-exposed proteins were also identified using in vivo phage display (Table 2). Some of these encode proteins that encode secretion signals but that are expected to be in the periplasm, e.g. the oligopeptide permease A2 and A4 subunits, and the flagellar subunit FlaB. Most are predicted to be cytoplasmic enzymes involved in the essential cell functions, such as glycolysis, transcription and translation. Several of the latter group, including BB0337 (enolase), BB0388 (RpoC) and BBG21 (hypothetical protein), have been found to have a non-specific ‘stickiness’, as they bind to plastic wells (L.T. Hu, pers. comm.). The others have not been tested, but some may display the same property. One interesting note, however, is that enolase has been reported to be localized on the surface of streptococci, where it binds plasmin(ogen) (Pancholi and Fischetti, 1998; Bergmann et al., 2003).
Table 2. Proteins not known or predicted to be normally localized on the outer surface of B. burgdorferi selected more than four times by in vivo phage display.
The name appears as annotated in the genome site at TIGR, but the functions of most of these proteins have not been demonstrated experimentally in B. burgdorferi.
1, 2, 3
4 heart, 3 joint, 1 bladder
1, 2, 3
13 heart, 7 joint
5 heart, 2 joint
5 heart, 2 joint
Oligopeptide permease A2
1, 2, 3
6 heart, 3 joint
1, 2, 3
8 heart, 3 joint
1, 2, 3
7 heart, 6 joint, 1 bladder
1, 2, 3
3 heart, 2 joint
1, 2, 3
5 heart, 1 joint
1, 2, 3
8 heart, 1 joint
1, 2, 3
4 heart, 3 joint
Oligopeptide permease A4
2 heart, 3 joint
Conserved hypothetical protein
1, 2, 3
4 heart, 2 joint
Conserved hypothetical protein
Cell binding activities of selected phage clones and MBP fusion proteins
To verify that selected clones encoding fragments of known or predicted surface proteins were not non-specifically ‘sticky’, we assessed attachment of representative phages to a number of mammalian endothelial and epithelial cell lines cultured in vitro. Cell lines were used because of the difficulties in obtaining primary murine endothelial cells in quantities sufficient for our experiments, and because primary cells are often not stable in in vitro culture. Phage clones containing in vivo-selected sequences from OspC, ErpK and Vls, bound to several mammalian cell lines (Fig. 4 and data not shown), as did P66-containing phage (Coburn et al., 1999; and data not shown). In multiple experiments, OspC, Vls and ErpK containing phage bound most efficiently to the A549 and LA4 cells, which are human and murine lung epithelial cell lines respectively. However, while OspC and ErpK containing phage did not bind to Ea.hy926 cells, the Vls containing phage did, supporting the hypothesis that different receptors are recognized. The vector phage, fdDog, also bound efficiently to the Ea.hy926 and 293 cells, but not to the A549 and LA4 cell lines, demonstrating that binding to the latter two cell lines is not simply conferred by the phage vector sequences. It is important to note here that binding to epithelial cells in vitro does not necessarily suggest that binding to endothelial cells in vivo is artefactual. Some classes of receptors, e.g. glycosaminoglycans, are expressed by multiple cell types, and not all receptors expressed by a particular cell type in vivo may be expressed by derivatives cultured in vitro.
To further ensure that the attachment of the B. burgdorferi-encoded protein fragments was not an artefact of expression in the bacteriophage particle, MBP fusions to ErpK and OspC were generated and tested for cell binding activity. These fusions also bound most efficiently to the A549 and LA4 cell lines (Fig. 5 and data not shown). Surprisingly, the BB0210 phage did not reproducibly or efficiently bind to any cell line tested, suggesting that either the apparent selection in vivo was an artefact, possibly due to the presence of several copies of the repeat sequence in close proximity on the phage particle, or perhaps because the receptor for BB0210 is not widely expressed by cultured cell lines.
Titring of individual phage clones in mouse tissues
Some of the in vivo-selected phage clones were enriched in particular tissues. To determine whether this was a result of tissue-specific homing of particular B. burgdorferi protein fragments, or was the result of a bias in the first round of selection that was amplified in the second and third rounds, we established tissue-specific titres of particular phage clones after injection into the tail veins of mice. Not surprisingly, all phage clones tested, including the vector, showed high titres in the spleen (Fig. 6). This is most likely a reflection of the circulatory system in that organ, as the vector and unselected library pools had the highest phage titres in the spleen (Table S1). In the remaining tissues, however, the titres of all clones were lower, and the vector control phage was clearly less efficient at binding to the endothelium than was any other phage clone. This result in itself is remarkable, as the vector control phage is more stable in vitro and in vivo than are phage containing large inserts, and are approximately 23-fold more infectious for Escherichia coli than are the library phage (data not shown). In all tissues, the P66 phage bound significantly more efficiently than did the vector phage (Fig. 6), and ErpK and Vls bound significantly more efficiently than the vector control in the ear (skin). Surprisingly, neither the BB0210 nor the OspC phage clone showed more efficient binding than did the vector phage in any tissue. This result, however, cannot be attributed to artefactual amplification in the first round of selection in any starting pool, because both were selected from all three starting pools of the phage library. While these results were somewhat different from the frequencies in which these clones appeared in the tissue-selected pools, they support the hypothesis that several particular B. burgdorferi protein fragments may have roles in binding of the bacteria to different mammalian receptors.
Summary and perspectives
The use of in vivo phage display has confirmed the adhesion activity of P66, the B. burgdorferi ligand for β3-chain integrins, in a physiologically relevant setting. In vivo phage display also identified several new candidate adhesins of the Lyme disease agent: OspC, Vls, certain Erp proteins and BmpD. One curious result of our in vivo phage display selection is the preponderance of B. burgdorferi proteins that were selected in the heart more frequently than in the other two tissues (Table 1), including BmpD and OspC. A trivial explanation is that, for whatever reason, more phages are randomly trapped within the heart than within other tissues, perhaps the result of the heart being the first organ harvested in this study in which the phages are exposed to arterial circulation after injection through the tail vein. This speculation is supported by our observation that the heart is efficient at trapping the vector and the unselected library pools (Table S1). However, because we required multiple independent isolations for any gene to be considered ‘selected’, and we did not deliver the phage library pools directly into the arterial circulation by intracardiac injection, other explanations are worthy of consideration. Although no other data are available for BmpD, our results for OspC are consistent with those published by other groups, which showed that ospC transcripts are more abundant in the heart than in any other tissue in B. burgdorferi-infected mice, although the adaptive immune response reduces ospC mRNA levels (Hodzic et al., 2003; Liang et al., 2004b). One model that is consistent with these results and others (Grimm et al., 2004b) is that OspC serves as an adhesin that is required for initial colonization of the mammal, and that B. burgdorferi cells that continue to express OspC after inoculation by the tick bind more efficiently to the endothelium in heart tissues during the dissemination of the bacteria. OspC has also been shown to bind to the tick salivary protein Salp15, which is believed to facilitate survival of the bacteria after inoculation into the mammal (Anguita et al., 2002). Because Salp15 is predicted to be heavily glycosylated (Anguita et al., 2002), it is possible that OspC recognizes glycoproteins of both tick and mammalian origin.
A second curious aspect of this work is that, aside from P66, we did not obtain any in vivo-selected clones encoding B. burgdorferi adhesins that have been characterized in vitro. These include the Borrelia glycosaminoglycan binding protein Bgp (BB0588), the fibronectin binding protein (BBK32), and the two decorin binding proteins, DbpA and DbpB (BBA24 and BBA25). It is likely that, within the lumen of a blood vessel, decorin is not available, as this proteoglycan is primarily associated with collagen fibrils in connective tissue. It is also entirely possible that serum proteins, e.g. fibronectin, removed the BBK32-containing phage from the pools, rendering them unavailable to adhere to the endothelium due to competition from the soluble protein. Another possibility is that dbpA/B and bbk32 were absent from, or under-represented in, our library. To address the possibilities of under-representation or absence, we performed quantitative polymerase chain reaction (PCR) on the starting library and on the selected pools (Fig. 7 and data not shown). While dbpA sequences were easily detectable in the starting library pools, this locus was undetectable in the in vivo-selected pools. We were unable to find a primer set that reliably amplified bbk32 from the genomic DNA of our starting B. burgdorferi clone, N40 D10E9. It is possible that this gene might be under-represented in the DNA from which the library was made, as bbk32 is not absolutely required for mammalian infection [Li et al. (2006) claimed that bbk32 is not required, while Seshu et al. (2006) showed that bbk32 mutants are attenuated]. Most but not all of the bacterial cells in the population may have lost the plasmid carrying the gene. A more likely possibility is that the bbk32 sequence in our N40 clone D10E9 may differ from published sequences from strain B31 and a different clone of N40 (Probert and Johnson, 1998), and therefore was inefficiently amplified by the PCR. That clone of N40 differs considerably from the one used in this work, as judged by the ospC and bbk2.10/ospF sequences.
The portion of the Bgp gene required for binding has not yet been identified, and the full-length gene is unlikely either to be efficiently amplified by quantitative PCR or to be present in the library, due to the nature of filamentous phage libraries (see below). There was evidence of selection of OspC, BB0210 and P66 in the selected pools, but we saw no evidence that bb0210 and ospC (bbb19) were over-represented in the library, although p66 (bb0603) was over-represented (data not shown). Analyses of the tissue-selected pools indicated that bb0210, bbb19 (ospC) and bb0603 (p66) were present at somewhat different levels in the selected pools (Fig. 7). These data demonstrate that the in vivo phage display approach was selective for particular genes in particular tissues, as ospC was more abundant in five of the six heart-selected pools than in two of the three joint-selected pools. In contrast, p66 was more abundant in the bladder-selected pool and two of the three joint selected pools than in any of the heart-selected pools. It is apparent, however, that selections in the same tissues were not always reproducible, a result that is likely a reflection of both selective and stochastic events that may occur in any selection. This emphasizes the importance of starting with multiple library pools and with multiple replicate selective media, in this case, mice.
There are a number of caveats to any filamentous phage display selection that are inherent to the biology of the bacteriophage itself. For example, the fusion of foreign sequences with those of phage gene III, as in the case of our library of B. burgdorferi DNA, requires that the open reading frame continue from the 5′ end of gene III through inserted sequence and through to the 3′ end of gene III. This would preclude the identification of proteins (and their corresponding genes) whose activities reside near or at the C-terminus of the protein. It is also possible that expression on the surface of a bacteriophage particle results in improper folding of the fusion protein, resulting in loss of adhesion activity. Furthermore, this approach is not applicable to the identification of proteins that are important for bacterial dissemination through tissue matrices, colonization of extravascular sites, or persistence in the host. This is due to the limited lifetime and tissue penetration of bacteriophage particles circulating in animals.
Phage display has previously been used to identify peptides that bind to the endothelium in living animals, and has more recently been used for the same purpose in humans (Pasqualini and Ruoslahti, 1996; Rajotte et al., 1998; Arap et al., 2002). This approach has demonstrated that the endothelium differs between tissues. To our knowledge, however, in vivo phage display selection from bacterial genomic libraries for proteins or peptides that serve as adhesins in a living animal has not been reported previously. The ease of genetic manipulations in some bacteria would obviate the need for approaches such as in vivo phage display, and for many pathogens, the sites of colonization would render this approach untenable or irrelevant. Even for B. burgdorferi, which does disseminate through the vascular system, in vivo phage display would not identify the bacterial molecules that are involved only in extravascular interactions, which is likely to be the case for the decorin binding proteins. Nevertheless, this approach should be amenable to use in the identification of pathogen adhesins that mediate interactions with mammalian receptors expressed in vivo, particularly when the pathogen cannot be grown in the laboratory or genetically manipulated with ease or confidence.
Bacterial strains, fd library and culture conditions
The filamentous phage display library of B. burgdorferi strain N40 DNA and associated E. coli growth and infection protocols were described previously (Coburn et al., 1999). The total genomic B. burgdorferi DNA used to make the library was isolated from an infectious clone (D10E9) of the tick isolate N40. The DNA was digested with Mse1 and Csp61, both of which leave 5′TA overhangs, under conditions that result in a light partial digestion of the intact DNA. Fragments between 200 and 1000 bp were purified from an agarose gel and ligated to adaptor oligonucleotides to allow ligation into the phage vector. The filamentous phage vector was derived from fd-tet (McCafferty et al., 1990; Clackson et al., 1991; Hoogenboom et al., 1991). The library host E. coli strain was MC1061 [F–araD139Δ(ara-leu)7696 galE15 galK16Δ(lac)X74 rpsL (StrR) hsdR2 (rk-mk+) mcrA mcrB1]. The E. coli strain TG1 [F+Δ(lac-pro) supE thi hsd5/F′traD36 proA+B+lacIqlacZΔM15], which expresses the F pilus required for phage fd infection, was used to amplify selected phage. The library was originally frozen down in six independent pools, but for the in vivo selections, the original pools were combined in sets of two (1 + 2, 3 + 4, 5 + 6), so that selections were performed with three independent starting library pools. Where required, media were supplemented with 12.5 μg ml−1 tetracycline to select for the presence of the phage genome. Aprotinin was added to 10−2 TIU ml−1 (trypsin inhibitory units ml−1) and benzamidine HCL was added to 1 mM to liquid cultures from which phages were harvested. After overnight growth of cultures at 37°C with agitation, phage particles were precipitated from 0.45 μm filtered culture supernatants with PEG/NaCl, as described (Coburn et al., 1999). For injection into mice, phages from 60 ml culture supernatant were resuspended in 200 μl HEPES-buffered saline (25 mM HEPES pH 7.8, 150 mM NaCl) supplemented with aprotinin and benzamidine (HBSAB).
In vivo selections of phage
A schematic depiction of the protocol is shown in Fig. 1. All protocols were approved by the institutional IACUC, and the mice were housed in the laboratory animal facility at Tufts-New England Medical Center. Female C3H/HeJ mice were used for all in vivo phage selections. This mouse strain was chosen because it is non-responsive to lipopolysaccharide (LPS) and therefore would not suffer unintended consequences in response to any LPS that might be present in the phage preparations. The mice were anesthetized with 2.5% avertin (2,2,2-tribromoethanol dissolved in one part tert-amyl alcohol; Papaioannou and Fox, 1993) in phosphate-buffered saline. Aliquots of the phage preparations were taken to determine the input titres. The mice were injected through the tail vein with 50–100 μl phage (at minimum, 1–5 × 1010 total phage), which were allowed to circulate for 4 min (Rajotte et al., 1998), after which they were sacrificed by cervical dislocation. The mice were then perfused through the left ventricle of the heart with 4 ml HBSAB. The outlet for perfusion was the right ventricle of the heart (the heart tissue would have been perfused because the coronary arteries branch off the aorta immediately after it emerges from the left ventricle). Note that for all phage preparations, the numbers of phage particles actually used could not be determined in advance of the experiment. This was due to the differential stability of different phage clones (J. Coburn, unpub. data), combined with the requirement for infection of E. coli followed by overnight growth of the transduced E. coli cells on selective medium. In the initial round of selection, the heart, tibiotarsal joints and urinary bladder were harvested from each mouse, rinsed with HBSAB, transferred to preweighed tubes containing 0.5 ml HBSAB and placed on dry ice. Although B. burgdorferi is known to infect skin as well, selection of phage in this tissue is problematic (Rajotte et al., 1998) and so was not performed. In subsequent rounds of selection, only the tissue from which the input phage pool had been selected in the previous round was harvested. After weighing, the tissues were homogenized by first chopping on dry ice, then grinding with glass mortars and pestles or disposable pestles in microfuge tubes, depending on the tissue. The tissue homogenates or dilutions thereof were used to infect naïve E. coli TG1, followed by growth on plates containing tetracycline. Phage pools were analysed after three rounds of selection in the same tissue. After the third round, the pools were injected into mice from which the heart, tibiotarsal joints, bladder, spleen and kidney were harvested to measure phage titres in the tissue in which that pool of phage had been selected versus other tissues. For a phage pool to be considered ‘tissue-selected’, the phage titre in the selective tissue had to be higher than those in all other tissues, including the spleen, which entraps phage efficiently. For example, for a pool to be considered ‘heart-selected’, it must have been selected in the heart three times, with the final selection round yielding a higher phage titre in the heart than in all other tissues measured. Only the ‘tissue-selected’ phage pools were analysed further.
Analysis of tissue-selected phage clones
Each tissue-selected pool was plated to obtain single colonies, which were restreaked to isolate pure clones. Each clone was first analysed by colony PCR to ensure the presence of B. burgdorferi DNA inserts, as described previously (Coburn et al., 1999), using fd primers listed in Table S2. The PCR products that contained B. burgdorferi DNA were sequenced by the Tufts University DNA/Protein core facility. The sequences were then matched to the B. burgdorferi strain B31 M1 sequence on the TIGR website (http://cmr.tigr.org/tigr-scripts/CMR/GenomePage.cgi?org=gbb) (Fraser et al., 1997; Casjens et al., 2000). For the sake of simplicity and universality, genes identified in this work are referred to by the TIGR designation, with apologies to the authors who have named the same loci or the proteins they encode differently. Alignments were performed using clustal.
Several criteria were used to identify B. burgdorferi genes and their corresponding proteins that were selected for binding to the endothelium in vivo. First, evidence of multiple independent selections was required. Multiple independent selections of a particular gene were determined to be either those obtained from different starting (input) library pools, or those containing overlapping but non-identical fragments of the same gene, regardless of the input library pool. In addition, for a protein to serve as an adhesin, that protein must be localized on the bacterial surface, so clones representing surface proteins were of greatest interest in this study.
Titring phage clones in mouse tissues
Individual phage clones expressing a representative sequence of bb0210, vls, p66, erpK or ospC and the empty vector control phage fdDog, were prepared as described above for the library stocks. Female C3H/HeJ mice were anesthetized with isofluorane, and injected through the tail vein with 50–100 μl of the phage preparations (at least 108 phages). The phages were allowed to circulate for 4 min, after which the mice were perfused through the heart (left ventricle) with 30 ml of perfusion buffer under physiologic pressure (the outlet was the right ventricle). The heart, bladder, ear, tibiotarsus and spleen were collected from each mouse at the end of the perfusion, and were placed in preweighed tubes and put on dry ice immediately. After weighing, the tissues were minced and homogenized with plastic disposable pestles (or, for tibiotarsi, glass-in-glass homogenizers) in perfusion buffer supplemented with 100 mM PMSF. The homogenates were diluted in L broth, then added to E. coli TG1 cells. Phage-infected cells were selected on 12.5 μg ml−1 tetracycline plates. The preference of the individual phages for specific tissues was calculated as the log of the transducing units (TU) of output phage per gm of tissue/TU of input phage in the individual experiment. The experiments for each phage clone were repeated at least three times. GraphPad Prism software was used to graph and calculate the statistical significance of the phage per gram tissue. The non-parametric Mann–Whitney U-test was used to calculate the statistical significance of the binding of each phage clone in each tissue compared with the control phage, fdDog.
Generation of MBP fusion proteins
The in vivo-selected fragments of ospC (bbB19) and erpK (bbM38) were amplified by PCR using the primer sets listed in Table S2. The gene fragments were digested with BamHI and SalI, then cloned in pMalC2 (New England Biolabs) that had been digested with the same enzymes. Clones containing the correct sequence were used to generate recombinant MBP fusion proteins as directed by the manufacturer's protocol and as previously described (Coburn et al., 1999). All of the proteins were present in the soluble fraction of E. coli and were stable during incubations with cells (data not shown).
Binding of phage clones and MBP fusion proteins to mammalian cells in culture
Most mammalian cell lines were purchased from the ATCC and cultured in the ATCC-recommended medium. The Ea.hy926 cell line was a generous gift from Dr Cora Jean Edgell, and was cultured as described previously (Edgell et al., 1983). For phage binding assays, the cells were plated in 96-well tissue culture-treated plates and allowed to grow to confluence. Phage stocks prepared as described above were suspended in HBSC + BSA buffer (25 mM HEPES pH 7.5, 150 mM NaCl, 1 mM MgCl2, 1 mM MnCl2, 0.25 mM CaCl2, 0.01 TIU ml−1 aprotinin and 1 mM benzamidine) plus 1% BSA (Coburn et al., 1999), and added to the cells, or to control wells without cells. To quantify the phage added to the wells, equal amounts were added to wells without cells and immediately fixed, then processed in parallel. After 3 h at ambient temperature, the cell layers were washed with HBSC, then fixed with paraformaldehyde. Binding was quantified by ELISA using a rabbit anti-M13 antiserum (Abcam, Cambridge, MA), essentially as described (Coburn et al., 1999). A similar protocol was used to quantify binding of MBP fusion proteins to cultured mammalian cells, with the exception that incubations were carried out for 1.5 h and the primary antibody used was anti-MBP (New England Biolabs).
The presence of selected gene fragments in the tissue-selected pools and in the starting library pools was assessed by quantitative PCR using SYBR green PCR master mix (Qiagen, Hilden Germany and Valencia, CA, USA). All B. burgdorferi targets were amplified using 4.18 × 107 copies of the phage DNA per reaction (as determined by OD260), and normalized to phage vector sequences, for which the reactions contained 4.18 × 105 copies of the phage DNA (by OD260). The standard curves consisted of known quantities of plasmid DNA (phage vector) or genomic DNA (B. burgdorferi genes) to allow quantification of the copy number in each phage pool. The final primer concentrations used were 0.2 μM. All reactions were performed for 40 cycles in a final volume of 25 μl in 96-well plates in a Stratagene MX3000p real-time thermocycler, and the data were analysed using the accompanying software. The copy number of each target B. burgdorferi gene was normalized to the copy number of the vector phage in each pool. Specific primer sequences and reaction conditions for each target are listed in Table S3.
We thank Dr Renata Pasqualini for generously providing many helpful suggestions at the start of this work; Carla Cugini, Reshma Williams and Erica Vallon for technical assistance in sequencing phage clones; Maria Angela Tanudra and Stephanie Carroll for expert assistance in titring phage clones in mouse tissues; Richard Karas and Mark Aronovitz for helpful suggestions regarding perfusion through the heart; and Michael Fisher and Joan Mecsas for advice on statistical analyses. This work was funded by a Biomedical Science Grant from the Arthritis Foundation, and by NIAID Grants R21 AI-059192 and R01 AI-051407 from the NIH. This work was also supported by the Center for Gastroenterology Research on Absorptive and Secretory Processes at Tufts-New England Medical Center, PHS Grant 1 P30DK39428 awarded by NIDDK.