Conserved lysine residues in decorin binding proteins of Borrelia garinii are critical in adhesion to human brain microvascular endothelial cells

Lyme borreliosis is a tick‐borne disease caused by Borrelia burgdorferi sensu lato spirochetes (Lyme borreliae). When the disease affects the central nervous system, it is referred to as neuroborreliosis. In Europe, neuroborreliosis is most often caused by Borrelia garinii. Although it is known that in the host Lyme borreliae spread from the tick bite site to distant tissues via the blood vasculature, the adherence of Lyme borreliae to human brain microvascular endothelial cells has not been studied before. Decorin binding proteins are adhesins expressed on Lyme borreliae. They mediate the adhesion of Lyme borreliae to decorin and biglycan, and the lysine residues located in the binding site of decorin binding proteins are important to the binding activity. In this study, we show that lysine residues located in the canonical binding site can also be found in decorin binding proteins of Borrelia garinii, and that these lysines contribute to biglycan and decorin binding. Most importantly, we show that the lysine residues are crucial for the binding of Lyme borreliae to decorin and biglycan expressing human brain microvascular endothelial cells, which in turn suggests that they are involved in the pathogenesis of neuroborreliosis.

disease affecting several organs. The three most common genospecies causing LB are B. burgdorferi sensu stricto (Bbss), B. garinii (Bg), and B. afzelii (Ba). Bbss causes most of the LB cases in North America while in Europe Bg and Ba are the dominant genospecies. Bbss is mainly associated with arthritis, Ba with the skin disorder acrodermatitis chronica atrophicans and Bg is associated with neuroborreliosis (Steere et al., 2016).
Lyme borreliae spread in the host via the blood vasculature  and are able to bind vascular endothelial cells (Coburn et al., 1998(Coburn et al., , 2013Salo et al., 2016). Once Lyme borreliae make the initial adherence contact with the endothelium they are able to transmigrate across the endothelium and reach the distant tissues Norman et al., 2008). There are a few studies where it has been shown that Lyme borreliae are also able to transmigrate across the endothelial barrier of the brain microvasculature (Bista et al., 2020;Grab et al., 2005Grab et al., , 2009Nyarko et al., 2006;Pulzova et al., 2011), while no studies have been published on the initial adhesion of Lyme borreliae to human brain microvascular endothelial cells (HBMECs).
Vascular interactions and adherence are mediated through multiple adhesins located on the outer surface of Lyme borreliae (Brissette & Gaultney, 2014). Decorin binding proteins A and B (DbpA, DbpB; Dbps) are surface-localized adhesins of Lyme borreliae expressed during mammalian infection (Brissette & Gaultney, 2014;Lin et al., 2017). The Dbps mediate binding to the proteoglycans decorin and biglycan. Proteoglycans are glycosylated proteins with a wide range of functions in the extracellular matrix, and they bind to other proteins through electrostatic interactions mediated by the negatively charged glycosaminoglycan (GAG) chains on their surface (Gandhi & Mancera, 2008). Decorin is expressed, for example, in extracellular matrix, while biglycan, on the contrary, is expressed also on blood endothelial cells (Salo et al., 2016). Previously, we have shown that Dbps of Bg bind biglycan on endothelial cells under flow conditions that mimic the blood flow in the vasculature (Salo et al., 2016).
The amino acid sequences of DbpA are more heterogeneous between different Lyme borreliae genospecies and strains compared to sequences of DbpB, which are conserved (Heikkilä et al., 2002a(Heikkilä et al., , 2002bRoberts et al., 1998). However, the tertiary structures of mature Dbps are similar, consisting of five helices, a linker between helices 1 and 2, and a C-terminal tail which can vary in length . According to the published structures of Dbps of Bbss and DbpA of Bg, the Dbps share a common ligand binding site Wang, 2012), which is a basic patch formed by residues of helices 2 and 5 (lysines 82, 163, and 170) (Brown et al., 1999;Pikas et al., 2003). The site is referred to as the canonical GAG-binding site, and it has, in fact, been shown to be critical for decorin binding and infection of mice (Brown et al., 1999;Fortune et al., 2014; Morgan Pikas et al., 2003;Wang, 2012).
Our objective was to study whether the conserved lysine residues in DbpA and DbpB are present in Bg strains, to model the structures of the proteins and to evaluate whether the lysines in the Dbps of Bg are critical for binding to decorin and biglycan. Moreover, since Bg is known to be the main causative agent of neuroborreliosis in Europe, we wanted to analyze whether the Dbps (and the lysine residues) of Bg are important also in mediating adhesion of Lyme borreliae to human brain microvascular endothelial cells.

| Sequence alignment and construction of mutated recombinant proteins
To assess whether the Dbps of Bg SBK40 also have lysine residues in the same positions as the previously analyzed DbpA of Bbss 297 (Brown et al., 1999), the amino acid sequences of these proteins  Figure S1). The alignment revealed that K80 is conserved among all published sequences, whereas K78 and K82 are conserved in the majority of sequences. In the nonconserved sequences, K78 is mostly substituted for aspartate or glutamate and K82 for arginine or glutamine. In the DbpB sequences, K78 is substituted for aspartate and K82 for glutamine.
To analyze the role of these lysine residues in proteoglycan binding, we mutated K78/80/82 in DbpA of Bg SBK40 and K79 in DbpB of Bg SBK40 to alanine. Sequencing of the constructs showed that the site-directed mutagenesis was successful, and no other mutations had been formed. Wild-type (wt) and mutated recombinant Dbps were then expressed in Escherichia coli and purified for further analyses.

| Thermofluor analysis
Thermofluor assay was performed to evaluate whether the mutations affected the stability of the recombinant proteins. The analysis revealed that wt DbpA had a melting temperature (Tm) of 72.2°C, which was ca. 7 degrees higher than the melting temperature of mutated recombinant DbpA (Tm 65.3°C) (Supplementary Figure S2).
The wt recombinant DbpB and mutated DbpB, on the contrary, had similar melting temperatures (Tm 57.6°C and 56.3°C, respectively).

| Structural analysis
To study the structural impact of the mutations and to analyze the potential binding sites, homology models were created for the Bg SBK40 wt DbpA and DbpB proteins and for the mutants (DbpA K78A/K80A/ K82A, DbpB K79A). Homology modeling is a commonly used protein structure prediction method where the structure of an unknown protein is modeled based on the structure of a related protein (França, 2015). DbpA of Bg SBK40 shares 89.6% sequence identity with DbpA of Bg PBr ((PDB ID: 2mtd, ) while DbpB shares 61% sequence identity with DbpB of Bbss B31 (PDB ID: 2mvg, ) and these structures were, respectively, used as structural templates for modeling the Bg SBK40 DbpA and DbpB. The models were carefully evaluated through visual inspection and by quality evaluation servers. The RMSD (root mean square deviation) to the template structures were 0.49 Å and 0.51 Å for DbpA and DbpB, respectively. The ModFOLD global model quality score was above 0.5 for both DbpA and DbpB (score > 0.4 indicates complete and confident models) and the ProQ LG scores were 4.988 and 4.724 for DbpA and DbpB, respectively (values > 4 indicates extremely good models).
Bg SBK40 DbpA and DbpB have high amino acid sequence similarity with other Dbps, and thus, share the same overall fold as previously published Dbp structures Fortune et al., 2014;Wang, 2012), consisting of five alpha helices and a linker region between helices 1 and 2 ( Figure 2).
The linker in DbpA has a helical structure, whereas in DbpB, and in most other DbpA structures, the linker forms an unstructured loop ( Figure 2). In the DbpB structure, helix 5 is significantly shorter than in DbpA and the C-terminal end is, therefore, longer and more unstructured. DbpA K80 and DbpB K79 are located in the canonical binding site, whereas DbpA K78 and K82 are located on the opposite side of the protein (Figure 2). K80,K161,and K168 in DbpA and K79,K160,and R167 in DbpB correspond to the canonical GAG-binding residues.
GAGs are known to interact with Dbps through electrostatic interactions and the electrostatic surface potential of the proteins can, therefore, give a good indication of possible GAG-binding sites. The

| Binding of recombinant proteins to biglycan and decorin
The binding of recombinant wt and mutated DbpB to biglycan and decorin in stationary conditions was studied with Western blot analysis and using a microtiter plate assay. Both analyses showed that wt DbpB bound well to both of the proteoglycans while the mutated DbpB did not show detectable binding in Western blot (Figure 4a), and in a microtiter plate assay the mutated DbpB bound both biglycan and decorin significantly less than the wt protein (p < .05 for biglycan, p < .01 for decorin) (Figure 4b).
F I G U R E 1 Multiple sequence alignment used for homology modeling. The Bg SBK40 DbpA and DbpB proteins are aligned to the structural templates used for modeling, Bg PBr DbpA (PDB ID: 2mtd, ) and Bbss B31 DbpB (PDB ID: 2mvg, ). The extensively studied DbpA protein of Bbss 297 is also included. The secondary structures of the templates are shown above (2mtd) and below (2mvg) the alignment. Conserved residues are shown with red background and similar residues are shown in red and boxed. The mutated residues are shown in black boxes and their positions are indicated by blue asterisks [Colour figure can be viewed at wileyonlinelibrary.com] The binding of DbpA to proteoglycans under stationary conditions was not studied since we have previously shown that wt DbpA of Bg SBK40 binds these proteoglycans only under flow (Salo et al., 2016). Therefore, the binding of wt and mutated Dbps to biglycan and decorin was studied under flow conditions with Biacore. In this analysis, wt DbpA bound efficiently both biglycan and decorin F I G U R E 2 The GAG-binding sites in the 3D structures of Bg SBK40 DbpA and DbpB. (a) Bg SBK40 DbpA (pink) superimposed on the structural template (light blue) Bg PBr DbpA (PDB ID: 2mtd, ). The close-ups show the canonical binding site, consisting of residues K80, K161, and K168, and the second binding site located on the opposite side of the protein. The second binding site shows residues K78 and K82 and residues K174 and E177, which correspond to K173 and K176 in Bg PBr DbpA (see text for more details). (b) Bg SBK40 DbpB (green) superimposed on the structural template (light blue) Bbss B31 DbpB (PDB ID: 2mvg, ). The close ups show the canonical binding site, which consist of K79, K160, and R167, and the potential second binding site located at the C-terminal end (residues K181, K182, N183, and K184). The linker of DbpB lacks the helical structure seen in DbpA and its helix 5 is shorter than in DbpA, and the C-terminal end of DbpB thus forms a longer flexible loop. The labels of the mutated residues are in italics and those located on the opposite site of the protein are in grey [Colour figure can be viewed at wileyonlinelibrary.com] (Figure 4c,d). Wt DbpB also clearly bound both of these proteoglycans, although the binding was not as strong as with wt DbpA. In contrast, the mutated recombinant Dbps could not bind biglycan or decorin under flow conditions. Moreover, none of the studied Dbps bound bovine serum albumin (BSA), which served as a negative control ligand (data not shown).

| Characterization of Lyme borreliae strains expressing wt or mutated adhesins
To analyze whether the specific lysines of Dbps are critical also for the binding properties of Lyme borreliae cells, we constructed Lyme borreliae strains that either express wt Dbps of Bg SBK40 (JF105 + pME22/WTdbps), mutated Dbps (JF105 + pME22/ MUTdbps), or no Dbps (JF105 + pME22) using Bbss JF105 as a background strain. Plasmid contents of the constructed strains were checked to verify that the strains had not lost any plasmids in electroporation. All of the studied strains had identical plasmid compo- Table S1). In addition, they all contained the plasmid pME22 used in transformation.

sition (Supplementary
Moreover, to verify that the wt and mutated Dbps were ex- Proteinase K treatment diminished, although not totally abolished, the signal observed in Western blots indicating that the proteins were also surface-localized. Detection of flagella protein p41 (periplasmic location) was used as a control. Proteinase K treatment had no effect on the signal obtained with p41 antibody and, as expected, the bands were observed around 41-42 kDa in both of the experimental conditions.

| The role of conserved lysine residues in binding to biglycan, decorin, and human brain microvascular endothelial cells
The binding of the constructed JF105 strains to biotinylated biglycan and decorin was analyzed with a dot blot assay. The results showed that only the strain expressing wt Dbps was able to bind the pro-  (Figure 6c). Next, transformed F I G U R E 3 Electrostatic surface potential maps of the Bg SBK40 DbpA and DbpB models. The electrostatic surfaces were calculated with the APBS tool (Adaptive Poisson-Boltzmann Solver) in PyMOL and the color ranges from −2 (red) to 2 (blue). The lysines located on the surface of the Dbps create positively charged areas that are involved in ionic interactions with GAGs. Mutating the DbpA residues K78/K80/K82 and DbpB K79 to alanines significantly changes the surface potential of the protein and prevents GAG-binding [Colour figure can be viewed at wileyonlinelibrary.com] JF105 strains were allowed to interact with HBMECs grown on microtiter plate wells. Unbound bacteria were washed away and the adhered carboxyfluorescein diacetate succinimidyl ester (CFSE)-

| D ISCUSS I ON
Dbps are important virulence determinants for Lyme borreliae.
They are expressed in a mammalian host, and strains lacking the Dbps are dissemination defective in the mouse model of LB (Imai et al., 2013). In addition, Dbps are required for the development of joint manifestations (Salo et al., 2015). One striking feature of Dbps is that, depending on the Lyme borreliae genospecies, the proteins bind proteoglycans and glycosaminoglycans with different affinities, which seems to have an effect on the tissue tropism of the spirochetes (Benoit et al., 2011;Lin et al., 2014;Salo et al., 2011Salo et al., , 2016). The Dbps have been shown to have lysine residues that are important in binding to decorin and glycosaminoglycans (Brown et al., 1999;Fortune et al., 2014;Pikas et al., 2003;Wang, 2012).
In this study, we show that DbpA and DbpB of Bg SBK40 also carry these critical lysine residues. K80 in DbpA and K79 in However, GAGs can also interact through hydrogen bonds (Gandhi & Mancera, 2008), and thus, the interactions with the GAGs could likely be maintained to some degree despite these residue differences. Another possibility is that the second GAG-binding site formed by these residues is not found in all Dbps but is instead compensated for by residues in other positions, as it has been shown that Dbps can have several binding sites that are uniquely adapted to each protein .
When K78/80/82 in DbpA and K79 in DbpB are mutated to alanine, the resulting recombinant proteins almost completely lose their ability to bind decorin and biglycan as demonstrated by Western blot, microtiter plate assay, and a flow-based Biacore assay. The binding of wt and mutated recombinant DbpA was analyzed only with a flow-based method since we have previously shown that wt DbpA of Bg SBK40 is able to bind proteoglycans only under flow conditions (Salo et al., 2016). These results show that the mutated lysine residues of both DbpA and DbpB play a role in GAG-binding.
As GAGs bind mainly through electrostatic interactions, the loss of surface charge in the mutants, as shown by the electrostatic surface potential maps (Figure 3), can explain the reduced GAG-binding seen in the mutants.
Thermofluor analysis was performed to analyze the folding and stability of the mutated proteins in comparison with the wt proteins.

F I G U R E 5
The surface localization of Dbps and binding of Bbss JF105 strains to biglycan and decorin. (a) The expression of Dbps on the surface of Bbss JF105 was detected with Western blot. Bacteria were incubated either without proteinase K (0 µg/ml) or with proteinase K (100 µg/ml) and loaded on 10% NuPAGE Bis-Tris protein gels and blotted on a nitrocellulose membrane. The membrane was incubated with rabbit-anti-DbpA/BgSBK40 or rabbit-anti-DbpB/BgSBK40 immune serum or as a loading control with rabbit-anti flagellin antibody. Bound antibodies were detected with horseradish peroxidase-conjugated goat anti-rabbit IgG. (b) JF105 strains were allowed to adhere to nitrocellulose membrane. After blocking, the membrane was incubated with biotinylated decorin or biotinylated biglycan. Bound proteoglycans were detected with horseradish peroxidase-conjugated streptavidin F I G U R E 6 The expression of biglycan and decorin in HBMECs and the binding of JF105 strains to the cells. Proteoglycan expression was detected with flow cytometry by using monoclonal biglycan antibody clone 4E1-1G7 (a) or monoclonal decorin antibody clone 9XX (b). Mouse IgG1 and mouse IgG2a κ clone were used as isotype control antibodies, respectively. Goat anti-mouse IgG Alexa Fluor 488 was used as a secondary antibody. Also, as a control, samples treated with the secondary antibody only were analyzed. The fluorescence was detected using Accuri C6. (c) Mean fluorescence intensities were determined with C6 software and a t test was used for statistical analyses. The data are expressed as mean fluorescence intensity of three independent samples ± standard deviation. *p < .05, **p < .01, ***p < .001. (d) HBMECs were grown on 96-well plates for 3 to 4 days. 50 × 10 6 carboxyfluorescein diacetate succinimidyl ester (CFSE) labeled bacteria were allowed to adhere to the cells at RT for 1 hr. Unbound bacteria were washed, and the wells were imaged with EVOS FL Auto 2 Cell Imaging System. Bound bacteria were calculated in ImageJ and the data were analyzed with Kruskal-Wallis test (corrected with Steel-Dwass post hoc test) with JMP Pro 14. The data are expressed as bacteria/100 cells in three independent experiments ± standard deviation [Colour figure can be viewed at wileyonlinelibrary.com] In this method, the protein is denatured in the presence of a dye, which binds to the gradually exposed hydrophobic parts of the protein. By binding to the protein, the dye becomes fluorescent, which can be monitored. The analysis revealed that the melting temperature (Tm) of mutated DbpA was ca 7°C lower than that of the wt protein. This implicates that the mutant DbpA is not as stable as the wt protein. However, we assumed that the folding of the mutated protein was not dramatically altered since the difference between the Tm values was rather modest. We performed MD-simulations for the 3D models of wt and mutant Dbps, which showed that they remained stable throughout the simulation, indicating that the mutations do not affect the 3D fold of the protein. We were also able to show that mutated DbpA was expressed on the surface of Lyme borreliae, supporting our assumption that the tertiary structure of the mutated protein was intact. Furthermore, others have demonstrated that point mutations from lysine to alanine have not altered the tertiary structure of other Dbps (Brown et al., 1999;Fortune et al., 2014).
It has been shown by  that DbpA of Bg Pbr has two GAG binding sites. In DbpA of Bg Pbr, and K82, while the negatively charged E177 will not be able to form similar ionic interactions with GAG chains.
To study the importance of the two binding sites in DbpA, a triple mutant K78/80/82A was created. By mutating only K80, which is totally conserved in all studied Dbp sequences and crucial for GAG-binding (Brown et al., 1999), the GAG-binding might still have been mediated through the potential second binding site, consisting of K78, K80, and K174. The corresponding second binding site in Bg PBr DbpA has been shown to have the same affinity for GAGs as that of the canonical binding site . Thus, by mutating residues from both binding sites we ensured that we had a nonfunctional protein suitable for performing a comprehensive study of the role of Dbps of B. garinii in adherence to HBMECs.
According to , K173/ K176 are more important in the second binding site than K77/K81. This indicates that K174 does not significantly mediate binding of Bg SBK40 DbpA to proteoglycans/HBMECs and, thus, we can conclude that our strategy of mutating all three lysines 78, 80, and 82 was suitable for identifying the residues critical for DbpA adhesion to

HBMECs.
DbpB of Bbss B31 also has multiple GAG binding sites . Of these, the canonical binding site is only secondary while the long unstructured C-terminal tail of the protein is the primary GAG binding site. However, when we mutated only one amino acid (K79) located in the canonical binding site in DbpB of Bg SBK40, we could not detect any binding to either of the studied proteoglycans in Biacore or Western blot analysis and only weak binding to biglycan in the microtiter plate assay. Thus, in contrast to DbpB of Bbss B31, it seems that in DbpB of Bg SBK40, the canonical binding site plays a crucial role in the binding of proteoglycans. Further, the C-terminal tail seems not to be the primary proteoglycan binding domain even though, according to the amino acid sequence and the predicted structure, DbpB of Bg SBK40 has a long unstructured C-terminal tail rich in lysine residues like DbpB of Bbss B31. Three lysines and one asparagine are the last residues in the C-terminal tail of Bg SBK40 DbpB, whereas the corresponding residues in Bbss B31 DbpB are all lysines. As asparagine is not a positively charged residue that could form ionic interactions with the GAGs, this difference might contribute to the reduced importance of the C-terminal tail as a GAG-binding site in Bg SBK40 DbpB. Moreover, the canonical GAG-binding site in Bbss B31 consist of two lysines and one glutamine . Since glutamine is also a noncharged residue this further suggests that the lysine-rich C-terminal tail of Bbss B31 DbpB can compensate for the missing lysine in the canonical binding site. In Bg SBK40 DbpB, the canonical GAG-binding residues are two lysines and an arginine (K79, K160 and R167). However, as arginines have been shown to bind more strongly than lysines to and a more compact shape, like the helical linker in Bg SBK40 DbpA, gives a more exposed binding pocket and increases the binding affinity. The DbpB mutant linker, which can reach further down toward the binding site, might, therefore, in a similar manner partly cover the binding site. The MD-simulation also showed that the backbone residues near K79/A79 are slightly more flexible in the mutant. K79 in wt DbpB forms hydrogen bonds and electrostatic interactions with the surrounding residues and its position is, therefore, more fixed by these interactions, whereas A79 in mutant DbpB lacks these interactions and, thus, confers flexibility.

Bg is the most common cause of neuroborreliosis in Europe and
it is thought to spread from the initial tick feeding site in the skin to the central nervous system via the blood circulation Stanek & Strle, 2018). To reach the central nervous system, Lyme borreliae must first spread via the vasculature, adhere to HBMECs, and traverse across the blood-brain barrier. The adherence of Lyme borreliae to various endothelial cells like human umbilical vein endothelial cells has been studied before (Coburn et al., 2013;Leong et al., 1998;Salo et al., 2016). Interestingly, it has been demonstrated at protein level that DbpA of B. bavariensis is able to interact with HBMECs (Tkáčová et al., 2020). In the dura mater of mice, Lyme borreliae are associated with the vascular regions (Divan et al., 2018), but to our knowledge, there are no previous studies on the adhesion of Lyme borreliae to vascular endothelial cells originating from human brain.
In this study, we constructed JF105 Lyme borreliae strains expressing the wt or the mutated Dbps of Bg SBK40 on their surface to study the role of Dbps and the lysines in endothelial adhesion.
Loss-of-function JF105 strain was used as the background in cloning because we wanted to evaluate whether the Dbps are important in binding to HBMECs when they are expressed on an infectious strain. This strategy results in more wt-like bacteria than the strategy of using a noninfectious background strain. By using a loss-of-function strain, we were also able to evaluate the role of Dbps in relation to other functionally redundant adhesins expressed on Lyme borreliae which also might contribute to HBMEC binding. In accordance with this strategy, we show that the Dbps of Bg SBK40 are able to mediate adherence to HBMECs. Lyme borreliae strain lacking Dbps bound significantly less to the cells than the strain expressing wt Dbps. These differences were observed even though JF105 strain most likely expresses redundant adhesins shown to be involved in binding to GAGs and endothelial cells (e.g., BBK32 (Fischer et al., 2006)). On the contrary, the use of a loss-of-function background strain that expresses other adhesins on its surface might explain why also the strain without Dbps bound HBMECs to some extent. Redundant adhesins might also, in part, account for the relatively high variance seen in the binding experiments. To conclude, by using the loss-of-function background instead of cloning the adhesins into a noninfectious gain-of-function background (e.g., B. burgdorferi B313), we were able to demonstrate the critical role of Dbps in HBMEC binding since other adhesins, despite their possible redundant roles, did not hide the binding effect of Dbps.
Previously, we have not been able to detect expression of decorin neither in human umbilical vein endothelial cells nor in endothelial cells of skin, heart, and brain tissues (Salo et al., 2016). In contrast, according to flow cytometry analysis, HBMECs clearly expressed both biglycan and decorin and, thus, both of these proteoglycans in HBMECs most likely act as ligands for Dbps. We were also able to demonstrate that, in addition to being important in recombinant proteins, the lysine residues in Dbps are also important when the proteins are expressed on Lyme borreliae. JF105 strain expressing the Dbps with lysine to alanine point mutations showed only background level binding to the HBMECs. These results indicate that the Dbps (and the lysine residues) most likely have a role in the pathogenesis of neuroborreliosis. Of note, Lyme borreliae have been shown to infect the dura mater of mice where decorin is highly expressed (Casselli et al., 2020).
The role of DbpA and/or DbpB of Bbss N40 in binding to another neuronal cell line (C6 glial cells) has been studied before by using a gain-of-function strain as a background in cloning (Fischer et al., 2003). It was shown that DbpB, but not DbpA, is able to bind C6 glial cells. DbpA versus DbpB of Bbss N40 exhibit different GAGbinding properties which is considered to account for the differences seen in cell adhesion. In addition, Dbps of N40 were not able to mediate adherence to EAhy-926 endothelial cells. In our study, we showed that when both DbpA and B of Bg SBK40 are expressed on a lossof-function background strain, binding to HBMECs is more efficient than binding of a strain that does not express Dbps. Thus, Dbps of different genospecies and strains, indeed, seem to account for the tissue tropism. Previously, we have shown that Dbps of Bg SBK40 act in a flow tolerant manner which is an important feature for adhesins during hematogenous dissemination and supports our hypothesis that Dbps of Bg have a role in the pathogenesis of neuroborreliosis also in vivo (Salo et al., 2016). The Dbps used in this study originate from Bg SBK40 strain which has been isolated from the human skin. However, since there is no mouse model for neuroborreliosis, the role of Dbps of Bg SBK40 in neuroborreliosis in vivo remains to be confirmed.
As a conclusion, we demonstrate that the lysine residues in the canonical binding pockets of Dbps of Bg are crucial for binding to biglycan and decorin. We also show that these lysines are conserved among different Bg strains, underlining their importance in the function of the Dbps. Importantly, Dbps and the lysines contribute to binding of Lyme borreliae to HBMECs and are, thus, potentially critical factors in the pathogenesis of neuroborreliosis.

| Bacterial strains and human brain microvascular endothelial cells
Escherichia coli strain M15 (Qiagen, Hilden, Germany) was used for cloning and expression of Dbps (Heikkilä et al., 2002a(Heikkilä et al., , 2002b. E. coli DH5α was used to purify pME22 plasmid containing the dbpBA operon of B. garinii SBK40 (infectious strain both in mice and men (Cuellar et al., 2019;Heikkilä et al., 2002b)). E. coli XL10-Gold (Agilent, Santa Clara, CA) was used in cloning of site-directed mutants. All E. coli strains were cultured in Luria Bertani broth or agar with appropriate antibiotics at 37°C overnight. Lyme borreliae strain JF105 (Bbss B31 missing linear plasmid 25, ΔdbpBA; loss-of-function strain) used in the study was a kind gift from Assistant Professor Yi-Pin Lin (State University of New York at Albany, Albany, NY) and has been described elsewhere Weening et al., 2008). All Lyme borreliae strains were cultured in Barbour-Stoenner-Kelly II (BSK II) broth at 33°C. All experiments were performed when the bacteria were at a logarithmic growth phase. The growth phase and the number of Lyme borreliae was studied using a Neubauer counting chamber.
Briefly, the cells were grown on collagen coated flasks or plates in Complete Medium with serum (Cell systems, Kirkland, WA) supplemented with CultureBoost-R (Cell systems), penicillin-streptomycin (Sigma, Steinheim, Germany) and 4 µg/ml blasticidin S (Sigma) with 5% CO 2 at 33°C. The experiments were performed with cells grown for 3 to 4 days.
PyMOL (The PyMOL Molecular Graphics System, Version 2.0.6 Schrödinger, LLC) was used for analyses, visualization and for preparing figures. The alignment figure was prepared using ESPript 3.0 (Robert & Gouet, 2014).

Molecular dynamics (MD) simulations of the Dbp models were done with Desmond Molecular Dynamics System (2019-4) in
Maestro (Bowers et al., 2006). The proteins (DbpA, DbpA-A78/ A80/A82, DbpB, and DbpB-A79) were solvated using a TIP3P water model (Jorgensen et al., 1983) in an octahedral box with 20 Å distance between the solute surface atoms and the edge of the box. The systems were neutralized by addition of Clatoms.
The OPLS3e force field (Harder et al., 2016) was used, with a con-

| Construction of mutated recombinant proteins
Lysines 78/80/82 in DbpA of Bg SBK40 and lysine 79 in DbpB of Bg SBK40 were mutated to alanine with site-directed mutagenesis.
Wild-type (wt) Dbp-proteins of Bg SBK40 cloned into pQ E30 expression vector of E. coli are described elsewhere (Heikkilä et al., 2002a(Heikkilä et al., , 2002b. The expression plasmids were purified from E. coli M15 with QIAprep Spin Miniprep Kit (Qiagen) according to the manufacturer's instructions. Purified plasmids were used as templates in the mutagenesis reactions. Mutations were prepared with QuikChange II Site-Directed Mutagenesis kit (Agilent) according to the manufacturer's instructions with primers listed in Supplementary Table S2.
Mutations were verified by sequencing the dbp-genes. Resulting plasmids containing the mutations were then transformed back into E. coli M15. Recombinant proteins were induced and purified as described elsewhere (Salo et al., 2011) and the protein concentrations were analyzed with BCA-kit (Pierce, Thermo Fisher Scientific, Waltham, MA) according to the manufacturers' instructions.
Total of 5 µl of the mixture was added on the polymerase chain reaction plate with 20 µl of PBS. We used triplicate as well as controls with no protein. The measurements were carried out using Light Cycler 480 II equipment (Roche Life Science, Mannheim, Germany).
In the protocol, the temperature was raised step by step from 20°C to 99°C. For the analysis, the data were exported, the average of three triplicate wells was determined for each data point, the reference value subtracted, and the curves were visualized. The melting temperature (Tm) was determined as temperature where 50% of protein was denatured.

| Binding of DbpB proteins to decorin and biglycan under stationary conditions
Binding of wt DbpB and mutated DbpB/K79A to decorin and biglycan was studied under stationary conditions with Western blot and in a microtiter plate assay.
For Western blot analysis, DbpB proteins were loaded on 10% Bis-Tris gels (NuPAGE, Thermo Fisher Scientific) and run in MES buffer (Novex, Thermo Fisher Scientific). Then, the proteins were blotted on a 0.2 µm nitrocellulose membrane (Whatman, GE Healthcare, Chicago, IL) and blocked with 3% skim milk in trisbuffered saline with 0.05% Tween20 (TBST) at RT for 2 hr after which the blocking solution was removed by rinsing the membrane with TBST. Biotinylated decorin or biotinylated biglycan (10 µg/ ml, biotinylated with EZ-Link NHS-LC-Biotin (Pierce, Thermo Fisher Scientific) according to manufacturer's instructions) was then incubated with the membrane at 4°C overnight. Unbound proteoglycans were removed by washing the membranes three times with TBST. Bound proteoglycans were detected by incubating streptavidin-horseradish peroxidase (HRP; 0.5 µg/ml, Pierce, Thermo Fisher Scientific) with the membranes at RT for 1 hr.
Unbound streptavidin was removed by washing the membranes three times with TBST and the bound streptavidin-HRP was detected with WesternBright ECL HRP substrate (Advansta, San Jose, CA) and Odyssey Fc imaging system (Licor Biotechnology, Bad Homburg, Germany). As a loading control, one SDS-PAGE gel loaded with Dbps was stained with SimplyBlue safe stain (Thermo Fisher Scientific) and imaged with Odyssey Licor Fc.
For the microtiter plate assay, purified Dbps were coated on enhanced binding microtiter plates (Thermo Fisher Scientific) 10 µg/ml in PBS for 1 hr at 37°C after which the wells were blocked with 1% BSA in PBS at 37°C for 1 hr. A 100 nM of biotinylated decorin and biglycan were added to the wells in PBS + 0.05% Tween 20 (PBST) and incubated at 37°C for 1 hr. As a control, two wells were incu-

| Binding of Dbps to decorin and biglycan under flow
Binding of Dbps to decorin and biglycan under flow was studied with surface plasmon resonance assay using Biacore X (GE Healthcare) as described elsewhere (Salo et al., 2016). Briefly, Dbps (50 µg/ml in 10 mM sodium acetate buffer pH 5) were immobilized on CM5 sensor chips (GE Healthcare) by amine coupling (

| Transformation of Lyme borreliae strains
The plasmids pME22, pME22/WTdbps, and pME22/MUTdbps were extracted and purified with Plasmid Midi kit (Qiagen) from E. coli according to the manufacturer's protocol. 1 × 10 9 JF105 bacteria were centrifuged at 1,500g and washed with ice-cold electroporation buffer (0.27 M sucrose and 15% (v/v) glycerol) three times. Then, about 30 µg of DNA was electroporated into JF105 with 2.5 kV, 200 Ω, and 25 µFa. After incubating the transformed JF105 in BSK II without antibiotics at 33°C overnight, 300 µg/ml kanamycin and 50 µg/ml gentamycin were added to the medium. The transformed bacteria were maintained in 96-microtiter plates (200 µl/well) at 35°C with 5% CO 2 . The resulting colonies were inoculated into 6 ml of fresh BSK II with 300 µg/ml kanamycin and 50 µg/ml gentamycin.

| Plasmid profiling
The plasmid profiles of the transformed Lyme borreliae strains were assessed by multiplex PCR (Bunikis et al., 2011). The DNA of the transformed Lyme borreliae strains and their parent strain was extracted with High Pure PCR Template Preparation kit (Roche Life Science) according to the manufacturer's protocol. Then, the circular and linear plasmids were amplified as described. The plasmids were stained with Midori green dye (Nippon Genetics, Düren, Germany) on 2.5% agarose gel.

| Immunodetection
The expression of wt and mutated DbpA and DbpB, and flagellin (p41) was detected with Western blot. 50 × 10 6 bacteria were collected by centrifugation and resuspended in 75 µl of water and 25 µl of 40× NuPAGE Sample Buffer (Life Technologies). Samples were loaded to 10% NuPAGE Bis-Tris protein gels (Thermo Fisher Scientific) and electrophoresis was run with MES-buffer (Thermo Fisher Scientific) at 200 V for 1 hr. The proteins were blotted as described above and blocked with 5% skim milk in PBS at 4°C overnight. The membrane was incubated with rabbit-anti-DbpA/ BgSBK40 or rabbit-anti-DbpB/BgSBK40 immune serum (MedProbe/ BioNordika, Oslo, Norway) in 5% skim milk in PBST at 4°C overnight or rabbit-anti flagellin antibody (1:1,000; Aviva Systems Biology, San Diego, CA) in 5% skim milk in PBST at RT for 1 hr. After washing the membrane three times with PBST, the membrane was incubated with HRP-conjugated goat anti-rabbit IgG (1:5,000; Santa Cruz Biotechnologies, Santa Cruz, CA) in PBST at RT for 1 hr. After washing, the proteins were imaged as described above.

| Proteinase K-assay
JF105 bacteria were washed with PBS containing 5 mM MgCl 2 , diluted to 1 × 10 8 bacteria/ml, and incubated with 0 or 100 µg/ml proteinase K (Sigma) at RT for hour. The bacteria were washed with the aforementioned buffer before analyzing the bacterial lysate samples by Western blot as described above.

| Flow cytometry
The expression of biglycan and decorin was detected in HBMECs with flow cytometry. HBMECs were detached from culture dish with 0.1% trypsin-EDTA. The cells were then washed once with serum containing culture medium and fixed at 33°C for 5 min with 4% paraformaldehyde (Santa Cruz Biotechnology). The fixative was removed by washing the cells three times with PBS. Unspecific protein-protein interactions were blocked by treating the cells with 0.1 M glycine in PBS at RT for 5 min. Proteoglycan expression was detected using monoclonal biglycan antibody clone 4E1-1G7 (dilution 10 µg/ml; Abnova, Taipei, Taiwan) or monoclonal decorin antibody clone 9XX (dilution 10 µg/ml; Santa Cruz Biotechnology). Mouse IgG1 (1 µg; Caltag Laboratories, Carlsbad, CA) and mouse IgG2a κ clone eBM2a (1 µg; Thermo Fisher Scientific) were used as isotype control antibodies. Goat anti-mouse IgG Alexa Fluor 488 (1:1,000; Invitrogen, Thermo Fisher Scientific) was used as a secondary antibody. Also, as a control, samples treated with the secondary antibody only were analyzed. The fluorescence was detected using Accuri C6 (BD, Franklin Lakes, NJ). The data were analyzed with C6 software (BD) and a t test was used for statistical analyses (IBM SPSS Statistics, Armonk, NY).

| Adhesion to HBMECs
HBMECs grown on 96-well plates for 3 to 4 days were fixed with cold 200 µl acetone at RT for 15 min. The wells were washed twice with 200 µl of PBST for 10 min. The cell nuclei were stained with 4ʹ,6-diamidino-2-phenylindole (DAPI; Invitrogen, Thermo Fisher Scientific) diluted 1:2,000 in PBS at RT for 5 min (protected from light). The wells were washed with PBS. The F-actin of the cells was fluorescently stained with Alexa Fluor 568 phalloidin (Invitrogen, Thermo Fisher Scientific) diluted 1:40 in PBS at RT for 1 hr and washed as described before. 50 × 10 6 CFSE (Molecular probes, Eugene, OR) labeled bacteria were allowed to adhere to the cells at RT for 1 hr. The staining procedure of Lyme borreliae with CFSE has been described elsewhere (Tuominen-Gustafsson et al., 2006).
The unbound bacteria were washed, and the wells were imaged with EVOS FL Auto 2 Cell Imaging System (Invitrogen, Thermo Fisher Scientific). Bound bacteria were calculated in ImageJ and the data was analyzed with Kruskal-Wallis test (corrected with Steel-Dwass post hoc test) with JMP Pro 14 (SAS, Cary, NY). The data are expressed as Borrelia/100 cells.

ACK N OWLED G M ENTS
We thank Joona Räikkä and Tuula Rantasalo for technical assistance. We thank the bioinformatics (J.V. Lehtonen), translational