Searching for Borrelial T cell Epitopes Associated with Antibiotic-Refractory Lyme Arthritis

Abstract

Antibiotic-refractory Lyme arthritis is believed to result from an infection-induced autoimmune response triggered by the spirochete Borrelia burgdorferi (Bb). Disease susceptibility is associated with the HLA alleles DRB1*0101, 0401, 0402, 0404, 0405 and DRB5*0101, and all these MHC molecules bind the Bb epitope OspA_163-175. However, not all patients have a proliferative response to this epitope. To identify other possible Bb epitopes involved in this disease process, the algorithm TEPITOPE was used to scan 17 immunogenetic Bb proteins for potential T cell epitopes with a refractory arthritis-associated MHC binding profile, and the Bb proteome was searched for peptides with sequence homology to OspA_165-173. Sixteen promising T epitopes were identified and their MHC binding profiles to 13 MHC molecules were verified using in vitro MHC/peptide binding assays. One peptide, BBK32_392-404, had a strong refractory-arthritis associated MHC binding profile, and another GK_297-306 shared sequence homology to OspA_165-173. However, patient cells did not proliferate in response to either peptide making it highly unlikely they were involved in a refractory course. A comparison of the in silico and in vitro results revealed that TEPITOPE correctly predicted 74% of the in vitro binding peptides, but it incorrectly predicted that 44% of the in vitrononbinding peptides would bind. For a particular MHC molecule, concordance between the in silico and in vitro results varied anywhere between 33% and 100%. Therefore, while additional Bb epitopes may be involved in the development of antibiotic-refractory Lyme arthritis, recognition of OspA_163-175 remains the only known Bb epitope associated with this disease.

1. Introduction

Lyme arthritis, which is caused by the tick-borne spirochete, Borrelia burgdorferi (Bb), is characterized by intermittent or persistent arthritis in a few large joints, particularly the knee. (Steere, 2001; Steere et al., 1987) In most patients, the arthritis can be treated successfully with either a 1 or 2-month course of oral doxycycline or a 2 or 4-week course of intravenous (IV) ceftriaxone. (Dattwyler et al., 1988; Steere et al., 1994) In rare cases however, synovitis persists for months or even several years despite treatment with ≥2 months of oral antibiotics, ≥1 months of IV antibiotics, or usually both. (Steere and Angelis, 2006; Steere et al., 1994) Since PCR results for Bb DNA in joint fluid are almost always negative in the post-antibiotic period (Carlson et al., 1999; Nocton et al., 1994; Steere and Angelis, 2006) and since cellular and humoral immune responses to Bb antigens begin to decline soon after antibiotic treatment (Kannian et al., 2007a; Kannian et al., 2007b), antibiotic therapy appears to result in the nearly complete or total eradication of spirochetes from the joint, yet synovial inflammation still persists. This disease course, which has been termed antibiotic-refractory Lyme arthritis, is thought to be the result of an infection-induced autoimmune response. (Steere and Glickstein, 2004)

A common feature of many autoimmune diseases is a link with specific HLA genotypes. Antibiotic-refractory Lyme arthritis is associated with the six HLA-DR molecules (DRB1*0401, 0101, 0404, 0405, 0402 and DRB5*0101) that bind the Bb epitope outer-surface protein A_165-173 (OspA_165-173) (Steere et al., 2006) and with T cell reactivity with this epitope. (Chen et al., 1999) In contrast, four MHC molecules (DRB1*0301, 0801, 1101 and 1104) that do not bind this OspA epitope are associated with an antibiotic-responsive course. How T cell reactivity to OspA_163-175 might be involved in antibiotic-refractory Lyme arthritis is presently unknown. Molecular mimicry between the OspA epitope and a sequence in human LFA-1 was proposed.(Gross et al., 1998) However, in subsequent studies, the LFA-1 peptide, as well as other human peptides with OspA_165-173 sequence homology, acted as only weak agonists in a small number of patients with refractory arthritis (Drouin et al., 2008; Trollmo et al., 2001), suggesting that they are unlikely to be relevant autoantigens in this illness. Furthermore, in a recent analysis using T cell tetramer reagents, three of seven DRB1*0401-positive patients with refractory arthritis lacked detectable OspA_165-173-specific T cells or proliferative responses to this epitope, whereas three of six patients with antibiotic-responsive arthritis had these OspA responses. (Kannian et al., 2007b) Thus, reactivity with this epitope, by itself, cannot explain all cases of antibiotic-refractory Lyme arthritis.

An alternative hypothesis is that T cell recognition of OspA_165-173, as well as certain other Bb epitopes, induce high levels of pro-inflammatory cytokines, which lead to bystander activation of autoreactive T cells and persistent synovitis. (Benoist and Mathis, 2001) Consistent with this hypothesis, we recently reported that patients with antibiotic-refractory arthritis had significantly higher levels of T_H1 chemoattractants and cytokines in joint fluid, particularly CXCL9 and IFN-γ, than patients with antibiotic-responsive arthritis. (Shin et al., 2007) Moreover, in patients with refractory arthritis, the levels of these chemokines and cytokines in synovial fluid remained high or even increased in the post-antibiotic period, while cellular and humoral reactivity with borrelial proteins, including OspA, declined. However, thus far, no borrelial epitope, other than OspA_165-173, has been identified that has an antibiotic-refractory arthritis-associated DRB binding profile and induces a T cell response in patients with refractory arthritis.

Bioinformatics programs can be used to predict T cell epitopes and their binding potential to DRB molecules. For example, the algorithm TEPITOPE, which is based on in vitro MHC/peptide binding of designer peptides, can predict the binding potential of T cell epitopes to 25 different DRB molecules. (Hammer et al., 1994; Hammer et al., 1997; Sturniolo et al., 1999) However, the ability of this program to identify potential disease-associated epitopes with a specific MHC binding profile has not yet been tested.

In the current study, we used TEPITOPE to screen 17 known immunogenic B. burgorferi proteins for T cell epitopes that were predicted to bind preferentially to antibiotic-refractory arthritis-associated DRB molecules. In addition, we searched the B. burgdorferi proteome for proteins with sequence homology to OspA_165-173. Promising epitopes identified in silico were tested for actual binding in vitro to 13 available HLA-DR molecules, including the six associated with antibiotic-refractory arthritis. One of the epitopes identified by the algorithm, BBK32_392-404 (Bb fibronectin binding protein), was demonstrated to have an antibiotic-refractory arthritis-associated DRB binding profile in vitro, while GK_297-306 (Bb glycerol kinase) had a high level of sequence homology with OspA_165-173. However, whereas many patients' cells recognized the OspA epitope, patients' T cells did not have a proliferative response to either the BBK32 or GK peptides.

2. Materials and methods

2.1 Computer databases and binding algorithms

Bb protein sequences were identified using PubMed and accession numbers are given in Table 1. Using TEPITOPE 2000 as a primary screening tool (on line at www.vaccinome.com), putative T cell epitopes were identified in 17 known immunogenic Bb proteins, and their predicted binding by 13 HLA-DRB molecules was assessed. In addition, the complete Bb B31 proteome was searched for peptides that had at least four amino acid residues in common with OspA_165-173 (YVLEGTLTA) using blastp and the filter disabled (http://www.ncbi.nlm.nih.gov/sutils/genom_table.cgi?taxid=%20Environmental&taxidinf=environ_info&selectall).

Table 1. Borrelial proteins assessed for T cell epitopes associated with antibiotic-refractory Lyme arthritis.

Known immunogenetic borrelial proteins
		GenBank Accession Number
1.	Arthritis Related Protein (ARP)	AAL25643
2.	Borrelia Membrane Protein (BMPA)	AAG00584
3.	Decorin Binding Protein A (DBP A)	NP_045697.1
4.	Decorin Binding Protein B (DBP B)	NP_045698
5.	Fibronectin Binding Protein (BBK32)	NP_212481
6.	Outer Surface Protein A (OspA)	NP_045688
7.	Outer Surface Protein B (OspB)	NP_045689.1
8.	Outer Surface Protein C (OspC)	AAA16058
9.	Outer Surface Protein E (OspE)	AAA22959.1
10.	Outer Surface Protein F (OspF)	AAA22960.1
11.	Vmp-Like Sequence, Expressed (VlsE)	AAC45733
12.	p30 (Similar to Oligopeptide Permease)	AAC44656
13.	p37 (Flagellin A)	NP_045619
14.	p41 (Flagellin B)	NP_212281
15.	p58 (Oligopeptide ABC transporter)	NP_212463
16.	p66 (Outer Membrane Porin Protein)	NP_2112737
17.	p93 (Protoplasmic Cylinder Antigen)	CAA49829.1
Borrelial proteins with a peptide sequence similar to OspA165-173
18.	Glycerol kinase (GK)	NP_212375.1
19.	Hypothetical protein BB0208	NP_212342.1
20.	Hypothetical protein BBG22	NP_045482.1
21.	Hypothetical protein BB0751	NP_212885.1

2.2 HLA-DR molecules and MHC/peptide binding assays

For B. burgdorferi T cell epitopes with promising predicted MHC binding profiles, actual peptide binding was determined in vitro using the same 13 HLA-DRB molecules that were assessed in silico. Based on a previous analysis of DRB binding of OspA_163-175 to these 13 DRB molecules (Steere et al., 2006), an antibiotic-refractory arthritis-associated DRB binding profile was defined as one in which a given peptide was bound by the HLA-DRB1*0101, 0401, 0402, 0404, or 0405 molecule or by the DRB5*0101 or DRB1*1501 molecule, but not by the DRB1*0301, 1101 and 1104 molecules. Since antibiotic-refractory arthritis was not clearly associated with either the DRB1*0102 or the DRB1*0701/DRB4*0101 molecules, these molecules were not included in the definition.

MHC/peptide binding assays were performed by incubating recombinant DRB molecules overnight with biotinylated peptides as previously described.(Steere et al., 2006) The following day, the MHC/peptide complexes were captured on 96-well plates coated with the monoclonal antibody L243, a HLA-DR specific antibody (American Type Culture Collection, Rockville, MD). After washing to remove unbound peptides, europium-labeled streptavidin (Wallac/PerkinElmer) was added. The plate was washed a final time and fluorescence was measured with a Victor² 1420 multilabel counter (Wallac/PerkinElmer).

All positive control peptides gave markedly positive results. The sequences of the positive control peptides were 0102-PC, PKYVKLNALKLAT for the DRB1*01 molecules; influenza NSI_32-45, FLDRLRRDQRSLRG for 0301; GAD65_557I, NFIRMVMSNPAAT for the 04 molecules, 1501 and B4*0101; 0701-PC, TSLYVRTSSFVIVSI for 0701; 1104-PC, PKYVKLNKLKSAT for 1101 and 1104; and HA_307-319, PKYVKQNTLKLAT for B5*0101. OspA_165A, KGAVLEGTLTAEK was used as a negative control peptide with all MHC molecules except the 04 molecules in which 04-NC, VSAATRGSLQATV was used (data not shown).

2.3 Proliferation Assays

To determine whether patients with Lyme arthritis had T cell reactivity with implicated Bb peptides, cells were tested from 15 patients, 8 with antibiotic-refractory arthritis and 7 with antibiotic-responsive arthritis, who were seen during a 7-month period between March and September 2005. All patients participated in a study, “Immunity in Lyme Arthritis”, which was approved by the Human Investigations Committees at Massachusetts General Hospital. T cell proliferation assays using bulk lymphocytes were performed as previously described (Chen et al., 1999). Briefly, PBMC (2 × 10⁵ cells per well) were incubated with peptides (1 μM final concentration) in duplicate wells. After five days, 0.5 μCi of ³H-thymidine (PerkinElmer) was added and cells were harvested 16-18 hours later. Incorporation of ³H-thymidine was measured using a TopCount scintillation counter (PerkinElmer).

2.4 Comparing in silico with in vitro results

The MHC/peptide binding results predicted by TEPITOPE were compared with those obtained by in vitro MHC/peptide binding assays. First, the result of each in vitro assay was categorized as to whether the target peptide bound or was not bound by the DR molecule. Binding was defined as a fluorescent value greater than 3SD above the mean value of the negative control peptides (≥ 90,000 relative fluorescent units). The observed results were compared with the predictions obtained using three threshold settings: a) ≥ 3% - high stringency; ≥ 10% - low stringency or c) <10 – not predicted to bind. In addition, the in silico and in vitro results were compared for each MHC molecule using the TEPITOPE threshold settings: a) 3% - high stringency; b) 10% - low stringency; and c) >10% - not predicted to bind. The level of concordance between the two methods was then computed.

3. Results

3.1 Bb T cell epitopes predicted to have an antibiotic-refractory Lyme arthritis binding profile

To identify Bb epitopes, in addition to OspA_165-173, with an antibiotic-refractory associated DRB binding profile, 17 known immunogenic Bb proteins (Akin et al., 2001; Bacon et al., 2003; Dressler et al., 1993; Heikkila et al., 2003; Salazar et al., 2005) (Table 1) were screened with the bioinformatics program TEPITOPE. This program scans proteins for peptides predicted to bind one or more of 25 different HLA-DR molecules, and it ranks peptide binding according to stringency (threshold parameter), which can be set anywhere from 1 (the most stringent) to 10 (the least stringent). In our initial screen, we used a threshold of 3, as recommended the program's designers, to keep the number of false-positive results reasonably low.(Bian et al., 2003)

In this initial screen of the 17 known immunogenic Bb proteins, 43 epitopes were predicted to be presented by the HLA-DR*0401 and 0101 molecules, the two MHC molecules most strongly associated with antibiotic-refractory Lyme arthritis. (Steere et al., 2006) All but two of the proteins, OspB and VlsE, contained such peptides. When these 43 epitopes were further assessed to determine their predicted binding profile to 10 additional HLA-DR molecules, 35 peptides, including OspA_163-175, demonstrated a high probability of presentation by four or more refractory arthritis-associated DRB molecules.

In addition to the screening of known immunogenic Bb proteins, we searched the borrelial proteome for peptides with sequence homology with the OspA_165-173 epitope, which is currently the only borrelial peptide known to have a refractory arthritis-associated HLA-DR binding profile. From this search, peptides from four different proteins were identified with homology to four or more of the OspA_165-173 residues: glycerol kinase (GK_297-306) and three hypothetical proteins (Table 1). GK_297-306, a housekeeping enzyme also expressed in human cells, shared identity with OspA_165-173 at five of the nine core residues, including two T cell contact sites. In addition, three of the six refractory arthritis-associated DRB molecules were predicted to bind the GK_297-306 epitope. In contrast, two of the hypothetical proteins were predicted not to bind any of the refractory arthritis-associated MHC molecules due to poor P1 anchor residues, and the third had a lysine at the P9 position, which was likely to inhibit binding to the shallow, hydrophobic P9 pockets of the DRB1*0101 and 0401 molecules.(Hill et al., 1994)

3.2 In vitro MHC/peptide binding assays of Bb epitopes

Because the availability of sufficient quantities of recombinant MHC molecules was limiting, 15 of the 35 epitopes from known immunogenic proteins predicted to bind at least four of the six refractory arthritis-associated DRB molecules (Table 2), along with the GK_297-306 epitope (Table 3), were selected for in vitro MHC/peptide binding studies to 13 available HLA-DRB molecules. These 13 MHC molecules represent nearly 80% of the DRB molecules found in patients with Lyme arthritis. (Steere et al., 2006) Twelve of these 13 DRB molecules were also part of the TEPITOPE algorithm; the program does not predict binding to the DRB4*0101 molecule.

Table 2. MHC/peptide binding profiles of borrelial epitopes predicted by TEPITOPE.

										Antibiotic-Refractory Associated									Antibiotic-Responsive Associated
									MHC	0401	0101	0402	0405	0404	B5*0101	1501	0701	0102	0301	1101	1104
Borrelial Peptides
Osp A
p8-16a	I	G	L	I	L	A	L	I	A	2b	1	2	4	2	3	2	4	1	6	1	1
p165-173	Y	V	L	E	G	T	L	T	A	1	1	2	1	1	5	3	4	3	7	1	2
p177-185	V	V	K	E	G	T	V	T	L	1	2	6	4	3	<10	7	3	2	7	<10	<10
BBK32
p394-402	Y	V	F	I	K	N	Q	K	N	1	1	1	1	2	1	9	4	5	<10	4	<10
Osp C
p14-22	F	L	F	I	S	C	N	N	S	1	1	3	1	1	1	8	4	1	6	1	5
OspE
p14-22	F	I	L	I	G	A	C	K	I	2	1	6	2	3	1	6	1	1	6	1	4
OspF
p5-13	M	F	I	I	C	A	I	F	A	2	1	2	3	1	1	2	2	1	3	1	1
BMPA
p111-119	Y	A	I	I	D	P	I	Y	S	1	1	4	1	2	1	7	3	3	3	1	1
DBP A
p15-23	L	T	I	L	V	N	L	L	I	1	1	1	2	1	6	1	2	1	9	2	2
DBP B
p14-22	F	F	K	L	L	V	A	C	S	1	3	6	1	2	8	<10	<10	5	7	1	2
p41
p1-9	M	I	I	N	H	N	T	S	A	1	2	1	2	1	<10	3	<10	2	7	8	8
p304-312	V	V	A	A	T	T	N	S	I	1	2	1	1	1	4	3	1	2	2	5	4
p58
p287-295	Y	F	Y	A	F	N	T	H	I	2	1	4	1	3	5	7	2	5	<10	9	<10
p295-303	I	K	P	L	D	N	V	K	I	2	1	3	2	1	9	3	4	1	9	7	6
p93
p620-628	V	V	S	E	S	N	F	E	I	1	2	4	5	3	6	2	2	2	8	<10	<10

^aAmino acid residues from the full length protein assessed for MHC binding using the TEPITOPE algorithm.

^bThreshold setting where MHC binding was predicted by TEPITOPE. Lower numbers correspond to a high stringency prediction (lower rate of false-positives, but higher rate of false-negatives), while lower numbers correspond to a low stringency prediction (higher rate of false-positives and lower rate of false-negatives). The value “<10” indicates MHC binding was not predicted at any TEPITOPE theshold setting.

Table 3. MHC/peptide binding profiles predicted by TEPITOPE of borrelial epitopes with sequence homology to OspA_165-173.

										Antibiotic-Refractory Associated									Antibiotic-Responsive Associated
									MHC	0401	0101	0402	0405	0404	B5*0101	1501	0701	0102	0301	1101	1104
Borrelial Peptides
Osp A
p165-173	Y	V	L	E	G	T	L	T	A	1	1	2	1	1	5	3	4	3	7	1	2
Glycerol Kinase
p297-305	“	“	“	“	“	S	V	F	I	1	2	8	1	2	5	<10	3	6	4	8	<10
Hypothetical Protein BB0208
p296-304	“	“	I	“	K	“	I	“	K	1	2	2	1	1	1	9	5	6	5	1	3
Hypothetical Protein BBG22
p90-98	K	T	P	V	“	“	“	“	“	<10	<10	<10	<10	<10	<10	<10	<10	<10	<10	<10	<10
Hypothetical Protein BB0751
p79-88	T	E	“	“	“	“	“	“	K	<10	<10	<10	<10	<10	<10	<10	<10	<10	<10	<10	<10

The results of the binding assays with the OspA_163-175 peptide were the same as in our previous analysis. (Steere et al., 2006) Six DRB molecules, which are found more commonly in patients with antibiotic-refractory arthritis (DRB1*0401, 0101, 0404, 0405, 0402 and DRB5*0101), bound OspA_163-175, whereas three molecules, which are found more often in patients with antibiotic-responsive arthritis (DRB1*0301, 1101, and 1104), did not (Figure 1). Regarding the MHC molecules found nearly equally in both patient groups, the DRB1*0102 molecule bound OspA_163-175 weakly, whereas both members of the disequilibrium pair B1*0701/B4*0101 did not.

Figure 1. In vitro binding profiles of OspA_163-175 and self-peptides to 13 recombinant HLA-DRB molecules.

MHC molecules are aligned on the X-axis according to their association with antibiotic-refractory or antibiotic-responsive Lyme arthritis (DRB1*1501/B5*0101 and DRB1*0701/B4*0101 are in linkage disequilibrium). MHC molecules were incubated with labeled peptides and MHC/labeled-peptide complexes were detected by fluorescence. The bars represent mean relative fluorescence of duplicate wells and the error bars denote the standard deviation. The amino acid sequence for each peptide is shown with core amino acids capitalized. Only one peptide (BBK32_392-404), in addition to OspA_165-173, had a strong antibiotic-refractory arthritis-associated DRB binding profile. All positive control peptides gave markedly positive results (data not shown). The average fluorescence of the negative control peptides was below 30,000 units (data not shown).

Of the 15 newly tested borrelial epitopes, only one, fibronectin binding protein_392-404 (BBK32_392-404), had a strong refractory arthritis-associated DRB binding profile (Figure 1). As with the OspA peptide, the BBK32 peptide bound the DRB1*0101 molecule, the four 04 molecules and the B5*0101 molecule. Moreover, in contrast with the OspA peptide, limited binding was detected to both members of the DRB disequilibrium pair DRB1*0701/B4*0101, which is found slightly more often in patients with refractory arthritis (Steere et al., 2006). The only deviation from the antibiotic-refractory DRB profile was the binding of BBK32_392-404 to the DRB1*1101 molecule, which is found more often in antibiotic-responsive patients. Though OspA_163-175 and BBK32_392-404 peptides had very similar DRB binding profiles, their nine core amino acid sequences were quite dissimilar with seven non-conservative substitutions including four at MHC contact sites.

Several other epitopes came close to having an antibiotic-refractory arthritis-associated DRB binding profile, but lacked strong binding to one or more of the critical DRB molecules. For example, OspC_12-24 bound five of the six refractory arthritis-associated molecules, rather like the OspA peptide, but it did not bind well to the DRB1*0101 molecule. GK_297-306 showed strong to weak binding to four of the six refractory arthritis-associated MHC molecules, but it lacked binding to the DRB1*0402 and 0404 molecules. The remaining 10 borrelial peptides also lacked binding to at least one of the refractory arthritis-associated DRB molecules. OspE_12-24 and OspF_3-15 had low-level binding to all of the MHC molecules tested, suggesting that this binding pattern was non-specific. Thus, only OspA_165-173 and BBK32_392-404 bound well to all six antibiotic-refractory arthritis-associated DRB molecules.

3.3 Recognition of Bb epitopes by patients' T cells

Because availability of patients' T cells is limited and because these cells cannot be replenished, we determined the immunogenicity of only OspA_160-177, BBK32_392-404 and GK_295-307. First, fresh PBMC from patients with Lyme arthritis seen over a five month period, five with antibiotic-refractory arthritis and six with antibiotic-responsive arthritis were stimulated with OspA_160-177 and BBK32_392-404 in standard proliferation assays (Figure 2A). T cells from four of the five patients with refractory arthritis and three of the six patients with responsive arthritis had proliferative responses to OspA_160-177 that were at least two times above background. However, cells from none of the 11 patients proliferated in response to the BBK32 peptide. Furthermore, PBL from 10 normal control subjects did not proliferate in response to either antigen.

Figure 2. Testing of the OspA_160-177, BBK32_392-404 and GK_295-307 peptides with Lyme arthritis patients' T cells.

(A) T cell proliferation assays were performed using fresh PBMC from 11 Lyme arthritis patients and 13 normal control subjects. Patients' cells often recognized the OspA peptide, but they did not have reactivity above background with the BBK32 peptide. (B) Frozen cells from the same 11 patients tested in panel A, and frozen cells from 8 normal control subjects were stimulated with the GK peptide. While some patients had a proliferative response to the OspA peptide, none had a proliferative response above background to the GK peptide. For patient 3, the recorded proliferative response was 12,551 CPM. Peptides are denoted on the horizontal axis and the average CPM of duplicate wells on the vertical axis.

Because the GK_295-307 peptide was identified after the BBK32_392-404 peptide, frozen cells from the same 11 patients that had been tested for responses to the BBK32 peptide and frozen cells from eight normal control subjects were tested for reactivity with the GK peptide. While two of the patients had a proliferative response to the OspA peptide, none of the patients or control subjects recognized GK_295-307. Thus, neither the BBK32 peptide nor the GK peptide seemed likely to be involved in an antibiotic-refractory disease course.

3.4 Comparison of computer binding predictions with in vitro MHC/peptide binding results

To assess the predictive powers of TEPITOPE, we compared the in silico MHC/peptide binding results (Tables 2 & 3) with the in vitro results (Figure 1). (DRB4*0101 was not included in this analysis as it was not part of the TEPITOPE program). To perform this analysis, in vitro MHC/peptide binding had to be classified as positive or negative. Therefore, a positive response was defined as a fluorescent value greater than three standard deviations above the mean value obtained with the negative control peptides (≥ 90,000 relative fluorescent units). According to this criterion, 117 of the 192 in vitro MHC/peptide binding assays performed had positive results. When these results were compared with TEPITOPE's predictions at a threshold of 3 (which was used for the initial screen), the program correctly predicted 74% (87/117) of the binding peptides (Figure 3). At the least stringent threshold setting of 10, TEPITOPE correctly identified 97% (113/117) of the binding peptides. However, among the 75 in vitro assays in which peptide binding did not occur, TEPITOPE incorrectly predicted that 44% (33/75) of these peptides would bind at a threshold of 3 and 87% (65/75) at a threshold of 10.

Figure 3. Assessing the predictive power of TEPITOPE.

The MHC/peptide binding results predicted by the algorithm were compared directly to the in vitro MHC/peptide binding results for 12 of the MHC molecules. True-positives identified by the algorithm were expressed as the percentage of the in vitro binding peptides that were predicted in silico using the threshold setting 3, 6 and 10 (black bars). The false-positive rate was expressed as the percentage of in vitro nonbinding peptides that were predicted in silico using the threshold settings of 3, 6 and 10 (grey bars).

The predictive power of the algorithm was also analyzed individually for the 12 MHC molecules tested (Table 4). At a threshold setting of 3, concordance between the in vitro and in silico results ranged from 33% to 100%. TEPITOPE's predictions were highest (90 to 100% concordance) for the DRB1*0402, 0101, 0404 and B5*0101 molecules; intermediate (60 to 90% concordance) for the DRB1*0102, 0401, 1501, and 0405 molecules; and lowest (≤50% concordance) for the DRB1*1101, 1104, 0701 and 0301 molecules. Interestingly, although the DRB5*0101 molecule bound all six peptides that it was predicted to bind, it also bound two peptides it was not predicted to bind at any threshold setting. Thus, the predictive power of the virtual matrix varied among MHC molecules.

Table 4. Concordance between in silico and in vitro peptide/MHC binding for individual HLA-DR molecules^a.

	Number of peptides bound in vitro / Total number of peptides predicted to be bound by TEPITOPE (%)b			Number of peptides bound in vitro / Total number of peptides not predicted to bind by TEPITOPEc
Threshold Ranged		≤ 3	≤ 10	>10
Concordance	MHC
HIGH	0402	9/9 (100)	12/16 (75)
	B5*0101	6/6 (100)	12/14 (86)	2/2
	0101	15/16 (94)	15/16 (94)
	0404	14/15 (93)	14/16 (88)
INTERMEDIATE	0102	10/13 (77)	13/13 (100)
	0401	11/16 (69)	11/16 (69)
	1501	5/8 (63)	9/15 (60)	0/1
	0405	8/13 (62)	9/16 (56)
LOW	1101	4/8 (50)	5/14 (36)	0/2
	1104	3/6 (50)	5/11 (45)	1/5
	0701	3/9 (33)	5/14 (36)	1/2
	0301	1/3 (33)	3/14 (21)	0/2

^aFifteen of the sixteen peptides were selected based on their prediction to be bound by DRB1*0101 and DRB1*0401 using a threshold range of 3. The only exception was GK_297-306.

^bFrequency TEPITOPE correctly predicted in vitro peptide binding. Binding of the 16 peptides to each MHC molecule was assessed by TEPITOPE. Peptides predicted to bind were then evaluated for in vitro binding [bound in vitro / total predicted to be bound in silico (percents in parentheses)].

^cFrequency TEPITOPE failed to predict in vitro peptide binding. Peptides not predicted to bind MHC molecules by TEPITOPE, at any threshold setting, were evaluated for in vitro peptide binding [bound in vitro / total predicted not to be bound in silico (percents in parentheses)].

^dTEPITOPE threshold setting (stringency) used to predict binding.

4. Discussion

We previously showed a highly significant association between antibiotic-refractory Lyme arthritis and specific HLA-DR molecules that bind the borrelial epitope OspA_163-175. (Steere et al., 2006) However, since not all patients with antibiotic-refractory Lyme arthritis have reactivity with this peptide, we searched here for other stimulatory borrelial epitopes with an antibiotic-refractory arthritis-associated DRB binding profile.

The number of potential epitopes is enormous. The Institute of Genomic Research (TIGR) calculates that the Bb proteome contains 1,740 known or hypothetical proteins, and each protein may have multiple T cell epitopes. Thus, we restricted our search to epitopes in 17 Bb proteins, which included most of the spirochete's known immunogenic proteins and those epitopes with OspA_163-175 sequence homology. Still, even this more restricted search included more epitopes than would be practical to test using in vitro MHC/peptide binding and T cell proliferation assays. Therefore, to further focus the search, we used the computer algorithm TEPITOPE to predict the MHC-peptide binding patterns of these borrelial epitopes. Of the 35 epitopes identified by the algorithm to bind at least four of the six antibiotic-refractory arthritis-associated DRB molecules, including the highly associated DRB1*0101 and 0401 molecules, 16 were selected for further analysis.

Next, we determined the in vitro the MHC/peptide binding profiles of these 16 borrelial epitopes to13 different HLA-DRB molecules, which required a total of 208 individual binding assays. Of the 16 borrelial peptides tested, only one (BBK32_392-404) had a strong antibiotic-refractory arthritis-associated DRB binding profile. In fact, its binding profile correlated with the HLA data even better than that of OspA_165-173 (Steere et al., 2006), since the genetically linked DRB1*0701 and DRB4*0101 molecules, which are found slightly more frequently in patients with antibiotic-refractory arthritis, bound BBK32_292-404, but not OspA_165-173. Patients with early Lyme disease frequently have an antibody response to the BBK32 protein and higher antibody responses correlate with less severe disease. (Akin et al., 1999) However, at this later stage of the illness, it appears that Lyme arthritis patients lack reactivity with this particular BBK32 peptide.

In addition, we looked for borrelial peptides with partial sequence homology with OspA_165-173. One possible epitope (glycerol kinase_297-306) was identified. As a key enzyme that shuttles glucose into the glycolytic pathway, glycerol kinase is expressed in both bacterial and human cells making it an interesting candidate molecular mimic. However, in vitro MHC/peptide binding assays demonstrated that the Bb GK_297-306 peptide was bound by only four of the six MHC molecules associated with refractory Lyme arthritis, and it did not stimulate patients' T cells. Therefore, we concluded that neither the BBK32 epitope nor the GK epitope was likely to be important in the pathogenesis of antibiotic-refractory Lyme arthritis.

It is certainly possible that borrelial epitopes other than OspA_163-175 might have a pathogenic role in antibiotic-refractory Lyme arthritis, but they may not bind all of the DRB molecules in vitro that are associated with a refractory course. First, the in vitro binding assay may not be sensitive enough to detect low-level binding that is still sufficient to activate patients' T cells. For example, we previously found that a human homologue of OspA_165-173, called MAWDBP_276-288, did not bind in vitro to the DRB1*1501 molecule or the genetically linked DRB5*0101 molecule, but this peptide stimulated T cells in some patients with Lyme arthritis who expressed these molecules (Drouin et al., 2008). Second, if certain borrelial peptides trigger particularly high pro-inflammatory responses, or even serve as a molecular mimics, they might do so in association with some, but not all, of the refractory arthritis-associated DRB molecules. Ultimately, the critical test always will be whether a given peptide stimulates patients' T cells. However, because of the limited availability of patients' cells and the lack of an animal model of antibiotic-refractory Lyme arthritis, it is currently possible to do such testing with only a limited number of borrelial peptides.

In this study, we were able to compare the in silico results to the in vitro results over a wide range of MHC molecules. Interestingly, we found that the predictive powers of TEPITOPE varied significantly depending on the DRB molecule. For example, at a threshold setting of 3, in silico and in vitro results were 94% concordant with the DRB1*0101 molecule, yet only 69% concordant with the 0401 molecule. Consequently, our initial screen, which was predicated on TEPITOPE identifying peptides bound by both DRB1*0101 and 0401, may have overlooked some candidate T cell epitopes. In other studies, TEPITOPE has been used successfully to identify biologically active promiscuous T cell epitopes from a single protein or small family of proteins in the context of cancer (Manici et al., 1999; Sturniolo et al., 1999), infectious disease (Panigada et al., 2002), or autoimmunity.(Hammer et al., 1997) However, the algorithm seems less effective at identifying T cell epitopes with a specific MHC binding profile from a large array of proteins. Therefore, we concluded that in vitro MHC/peptide binding assays must still be performed to confirm peptide binding.

In summary, using a combination of the TEPITOPE program, in vitro MHC/peptide binding assays, and sequence alignment algorithms, we identified one new borrelial epitope (BBK32_392-404), in addition to OspA_163-175, that had a strong antibiotic-refractory arthritis-associated DRB binding profile, and another (GK_297-306) with sequence homology with the OspA epitope. However, T cells from patients with Lyme arthritis did not proliferate in response to either of these epitopes. Even so, other as yet unidentified borrelial epitopes may play a pathogenic role in the development of antibiotic-refractory Lyme arthritis.

Abbreviations

OspA

outer-surface protein A

Bb

Borrelia burgorferi

BBK32

fibronectin binding protein

GK

glycerol kinase