Human Autoreactive CD4+ T Cells from Naive CD45RA+ and Memory CD45RO+ Subsets Differ with Respect to Epitope Specificity and Functional Antigen Avidity

T cells with specificity for self-Ags are normally present in the peripheral blood, and, upon activation, may target tissue Ags and become involved in the pathogenesis of autoimmune processes. In multiple sclerosis, a demyelinating disease of the CNS, it is postulated that inflammatory damage is initiated by CD4+ T cells reactive to myelin Ags. To investigate the potential naive vs memory origin of circulating myelin-reactive cells, we have generated myelin basic protein (MBP)- and tetanus toxoid-specific T cell clones from CD45RA+/RO− and CD45RO+/RA− CD4+ T cell subsets from the peripheral blood of multiple sclerosis patients and controls. Our results show that 1) the response to MBP, different from that to TT, predominantly emerges from the CD45RA+ subset; 2) the reactivity to immunodominant MBP epitopes mostly resides in the CD45RA+ subset; 3) in each individual, the recognition of single MBP epitopes is skewed to either subset, with no overlap in the Ag fine specificity; and 4) in spite of a lower expression of costimulatory and adhesion molecules, CD45RA+ subset-derived clones recognize epitopes with higher functional Ag avidity. These findings point to a central role of the naive CD45RA+ T cell subset as the source for immunodominant, potentially pathogenic effector CD4+ T cell responses in humans.

I mmunological research on organ-specific autoimmune diseases has focused to a large extent on the identification of candidate self-Ags. In this context, myelin proteins have been studied as potential targets of autoimmune reactivity in the pathogenesis of multiple sclerosis (MS) 2 , the most common demyelinating disease of the CNS. Ags such as myelin basic protein (MBP) have been employed as immunogens in vivo to learn from experimental disease in animals (experimental allergic encephalomyelitis, EAE), and in vitro to study the human immune response in MS patients (reviewed in Ref. 1). EAE studies have demonstrated that inflammatory demyelinating CNS disorders can be mediated by CD4 ϩ myelin-specific T cells. Extensive data in humans, on the other hand, indicate that myelin-reactive cells are part of the normal T cell repertoire both in MS patients and healthy subjects (2)(3)(4)(5). However, these studies on the frequency of self-reactive T lymphocytes did not address whether these cells had been Ag primed in the body, i.e., whether they derived from a memory or naive pool. This is a relevant question, since naive and memory T cells play different roles in the immune response (6 -8) and differ with respect to their Ag presentation and activation requirements (9 -11), their TCR signaling (12), and their tissue-specific homing receptor-ligand interactions (13)(14)(15)(16).
Since human T cells upon antigenic priming down-regulate the high-m.w. CD45RA isoform and reciprocally up-regulate the lowm.w. form CD45RO, these molecules have been proposed as markers for naive (CD45RA ϩ /RO Ϫ ) and memory (CD45RA Ϫ / RO ϩ ) T cells (17,18). The following well-established observations corroborate this notion: 1) CD45RO ϩ cells are reduced under conditions of limited antigenic exposure, such as in human cord blood or in neonatal mice; 2) CD45RO ϩ cells are almost absent in the spleens of mice bred in a germfree environment; 3) increased proportions of CD45RO ϩ T cells are found in the peripheral blood of aging or immunized humans and animals; and 4) CD4 ϩ CD45RO ϩ T cells predominantly give rise to the human response to recall Ags such as tetanus toxoid (TT) in immunized subjects (19). More recent data, however, demonstrated that changes in CD45R isoforms are reversible in the rat (20,21). Since evidence of such reversion in humans is so far only indirect (22), CD45RA vs CD45RO expression has been widely accepted and used to identify and select human naive and memory CD4 ϩ T cells, respectively. A concomitant expression of both CD45RA and CD45RO has been associated with the phenotype of effector T cells, but the relationship of this cell population with that of naive and memory cells is not fully understood (8,23). Thus, we studied human myelin Ag-specific responses from CD4 ϩ T cell populations identified by clear expression of either the CD45RA or CD45RO isoform postulating they represent resting naive and memory T cells, and excluded double-positive cells which may either represent a transitional or an effector state.
To better understand the origin, the functional characteristics, and the specificity of CD4 ϩ MBP-specific T cells derived from the peripheral blood of MS patients and controls, we asked 1) whether autoreactive T cells can originate both from the putatively naive (CD45RA ϩ /RO Ϫ ) and memory (CD45RA Ϫ /RO ϩ ) CD4 ϩ subsets; 2) whether effector autoreactive cells raised from these subsets express different T helper phenotypes; and 3) whether the MBPreactive T cell repertoires originating from the two subsets differ with respect to frequency, epitope recognition, and antigenic affinity.

Subjects
Donors were typed for HLA class II using standardized molecular typing methods by the Department of Transfusion Medicine, National Institutes of Health. To include well-characterized backgrounds in terms of HLA-restricted T cell response to MBP, we selected patients with haplotypes including DRB1*1501, *1502, or *0401. Age, gender, and HLA-DR haplotype of the subjects were as follows: MS patients MS1, 31-

Ags
Human MBP was prepared according to the method of Deibler et al. (24). A single batch of MBP of known purity and in vitro immunogenicity was used throughout the entire study. MBP peptides were synthesized on a Beckman 990 peptide synthesizer (Beckman Coulter, Fullerton, CA) using Merrifield's solid-phase synthesis methodology according to published sequences (5). The superantigens (SAgs) staphylococcal enteroxin A, staphylococccal enteroxin B, staphylococcal enteroxin C-1, -2, -3, staphylococcal enteroxin D, staphylococcal enteroxin E, and toxic shock syndrome toxin-1 were obtained from Toxin Technologies (Sarasota, FL). TT was purchased from Pasteur Merieux Connaught (North York, Ontario, Canada).

Purification of CD4 ϩ T cell subsets
PBMC were obtained from leukoapheresis buffy coats by density gradient centrifugation (Organon Teknika, Durham, NC). T cell subsets were isolated from 1-2 ϫ 10 8 freshly obtained PBMC via high-affinity negative selection using human CD4 ϩ /45RO Ϫ or CD4 ϩ /45RA Ϫ subset column kits (R&D Systems, Minneapolis, MN), essentially according to the manufacturer's protocol. Our improvements included thorough resuspension of the cell pellet by gently passing the cells six times through a 16-gauge needle before loading and keeping a steady cell elution flow by carefully maintaining a 2-ml wash buffer volume in the column reservoir. Cryopreserved cells proved unsuitable for the separation procedure due to reduced viability and an increased tendency for aggregation. Therefore, only data obtained from subset separations of freshly collected PBMC were included in this study. Composition of the PBMC and purity of the separated T cell subsets were routinely checked by flow cytometry using a FACScan-and CellQuest software (Becton Dickinson, Mountain View, CA) after staining with fluorochrome-labeled anti-CD3, CD4, CD8, CD45RA, and CD45RO mAb as described below. The mean purity of all separations as determined by FACS staining with mAb specific for CD45RA and CD45RO was Ն95% for each subset. Double-positive (CD45RA ϩ /RO ϩ ) "effector" cells, variably present in the PBMC (9 and our observation), were selected out. Cell viability was always Ͼ98% by trypan blue staining exclusion. The yield of purified cells recovered from all cells loaded into the column ranged from 25 to 45%, in accordance to the manufacturer's specifications. To assess whether cell loss biased the phenotypic characteristics of T cell subsets, we performed FACS analysis of cell surface markers including, in addition to CD45 isoforms, costimulation receptors (CD28), cell adhesion molecules (CD11a, CD49d, and CD62L) and activation markers (CD25 and CD69). The expression of these molecules on the isolated CD45RA ϩ or RO ϩ subset reflected the expected phenotypes of naive and memory T cells and matched the profiles found on the corresponding subset analyzed by electronic gating in whole PBMC (data not shown). These experiments also confirmed that the negative selection strategy avoided functional alterations consequent to engagement of CD45 molecules, yielding unaltered cells.

Generation and characterization of Ag-specific T cell clones
T cell clones (TCC) specific for MBP were established from purified T cell subsets using a modification of the split-well method (2). From each donor, purified CD45RA ϩ /RO Ϫ and CD45 RA Ϫ /RO ϩ CD4 ϩ T cells (2.5 ϫ 10 4 / well) were seeded into 96-well plates (Nunc, Roskilde, Denmark) along with autologous irradiated PBMC (3000 rad, 1.5 ϫ 10 5 /well) as APC and stimulated with 30 g/ml MBP or 10 g/ml MBP peptide. Wells containing specific proliferating T cells (stimulation index Ͼ2 with a ␦ cpm Ͼ1000) were periodically (10 -15 days) restimulated with Ag and APC and expanded in medium containing 5 ng/ml of human rIL-2 (generously provided by Prof. Sebald, Biozentrum, Würzburg, Germany). Clonality of T cell cultures was assessed by PCR-assisted TCR V␤ gene usage analysis as described elsewhere (5).
Concentrations of the soluble factors IFN-␥, TNF-␣, IL-4, IL-10, and GM-CSF were measured by ELISA using CytoSets (BioSource, Camarillo, CA) according to an optimized standard protocol. Aliquots of cell culture supernatant (50 l) were harvested 48 h after Ag stimulation and stored at Ϫ80°C until analysis. Supernatants were diluted 1:5 with assay buffer and used in a total assay volume of 75 l/well. For each TCC tested, we measured cytokine secretion upon the following conditions: no Ag, 10 and 100 g/ml MBP, 10 and 100 g/ml specific MBP peptide. All measurements were performed in duplicate wells, and the specificity of proliferation was ascertained in every instance.
Antigenic peptide specificities of the MBP-reactive TCC were defined by using a panel of 16 overlapping 19-mers in standard proliferation assays. Epitopes within the immunodominant region of MBP(111-129) were further mapped using a set of 16 peptides truncated at the amino (112 to 119 -129) and carboxyl (111-121 to 128) termini at equimolar concentrations (5). The overall functional Ag affinity was estimated by determining the Ag concentration required to induce 50% of the maximum proliferation (EC 50 ). This was accomplished in dose-response proliferation assays covering Ag concentrations between 10 Ϫ5 and 10 2 g/ml.

Statistical methods
Given the difficulty in obtaining large numbers of purified T cells to set up full limiting dilution experiments, we chose culture conditions aimed to obtain MBP-specific T cell lines arising from a single precursor cell per well. Such expectation was confirmed by the clonal profiles of TCR usage of the T cell lines (data not shown). This allowed us to estimate T cell precursor frequency of MBP-responding cells by dividing the number of responding wells by the number of wells seeded, multiplied by the number of cells per well.
Variance was measured by calculating SD. Differences between groups were evaluated by the unpaired t test. For data not normally distributed, the Mann-Whitney U rank sum test was applied. Statistical calculations were performed using SigmaStat software (Jandel Scientific, San Rafael, CA).

Results
The human T cell response to MBP originates predominantly from the CD45RA ϩ CD4 ϩ T cell subset To estimate the frequency of autoantigen-specific T cells in the naive and memory resting CD4 ϩ T cell population, we obtained highly purified CD45RA ϩ /RO Ϫ and CD45RA Ϫ /RO ϩ cell subsets by negative selection. To assess the maximum proliferative capacity of the isolated subsets in response to potent stimuli, we compared the primary response to SAg (staphylococcal enteroxin A, staphylococcal enteroxin B, SEC1-SEC3, staphylococcal enteroxin D, staphylococcal enteroxin E, and toxic shock syndrome toxin-1) at concentrations ranging from 10 Ϫ2 to 10 2 ng/ml in doseresponse proliferation assays. Cells from both subsets proliferated comparably well, indicating that universal T cell stimuli elicit similar responses both in naive and memory T lymphocytes. In contrast, the short-term reaction to TT in primary proliferation assays clearly resided in the CD45RO ϩ subset, as expected from a response to a recall Ag in immunized subjects. However, although a vigorous primary proliferation to TT (assayed after 3 days) could be detected only in the CD45RO ϩ subset, short-term TT-specific cell lines could be raised after rechallenge with the Ag from the CD45RA ϩ subset as well (data not shown).
We next dissected the T cell response of the subsets to MBP using purified T cell subsets and APCs. Interestingly, it was the CD45RA ϩ T cell population that originated predominant responses to MBP, both in MS patients and in healthy controls (Table I). The estimated precursor frequency of MBP-reactive T cells in the CD45RA ϩ subset was significantly higher than in the CD45RO ϩ subset of MS patients, (6.10 Ϯ 3.41 ϫ 10 Ϫ6 vs 2.59 Ϯ 2.13 ϫ 10 Ϫ6 ; p ϭ 0.040). Although this was true also for healthy donors (7.24 Ϯ 1.79 ϫ 10 Ϫ6 vs 2.67 Ϯ 1.33 ϫ 10 Ϫ6 ), the small number of subjects available for analysis does not allow us to compare their precursor frequency of MBP-reactive cells with that of MS patients.

MBP-specific TCC originating from the CD45RA ϩ and CD45RO ϩ subsets recognize different antigenic epitopes within individual subjects
To assess the Ag-fine specificities of TCC originating from either subset, we tested MBP-reactive TCC in proliferation assays using a panel of 16 overlapping MBP peptides. As summarized in Fig. 1, the overall specificities corresponded well to the pattern of immunodominant MBP epitopes previously identified by others and by us  (Table II). Even in the case of naive-and memory-derived TCC from a given subject responding to the same peptide MBP (111-129) (patient MS4), further epitope mapping with truncated peptides delineated different nested epitopes, probably recognized by

Cytokine secretion profiles of CD45RA ϩ and CD45RO ϩ subset-originating TCC
To define the T helper subtype of MBP-specific T cells emerging from the CD45RA ϩ and CD45RO ϩ subsets of MS patients, we measured the secretion of pro (IFN-␥, TNF-␣)-and antiinflammatory cytokines (IL-4, IL-10) in culture supernatants of 41 TCC raised from 4 subjects , as well as the APC-activating factor GM-CSF. The conventional types of T helper cells (Th1, Th2, and Th0) emerged from both the naive and the memory subsets (Table  III). By using the IFN-␥ to IL-4 quotient as a marker of Th1-vs Th2-type cells, we failed to detect clear-cut systematic differences in the cytokine secretion of TCC derived from the naive and memory subsets (Fig. 2). Although the small number of memory subsetderived TCC available for analysis precludes a definitive statement on their phenotype, naive subset-originating TCC were predominantly Th1 or Th0, similar to observations on PBMC-derived TCC specific for immunodominant MBP epitopes (5,29).

Higher functional avidity of epitope recognition by CD45RA ϩ subset-derived TCC does not depend on higher expression of costimulatory/accessory molecules
To estimate the functional avidity for the Ag of naive vs memory subset-derived MBP-specific T cells, we determined the EC 50 of 35 representative, well-characterized TCC (25 naive and 10 memory derived). CD45RA ϩ subset-derived TCC had significantly lower EC 50 than CD45RO ϩ subset-derived TCC (0.50 vs 6.25 g/ml, median; p Ͻ 0.001, Fig. 3). Since the EC 50 measure does not represent the receptor/ligand affinity in physicochemical terms, but rather a functional equivalent of complex interactions involving a number of molecules, we examined the expression of the most important cofactors involved in cell proliferation in these TCC. Among molecules regulating signaling and homing of naive and memory T cells, we analyzed CD45RA, CD45RO, CD62L  Fig. 4. Interestingly, even after repeated in vitro Ag stimulation of the TCC, consistent differences in their CD45R isoform expression were still clearly detectable. The proportion of CD28 ϩ cells was greater in memory-derived TCC ( p ϭ 0.003). CD29 expression was also strongly associated with memory-derived TCC, both in terms of percent positive cells ( p Ͻ 0.001) and staining intensity ( p Ͻ 0.001). We observed a more prominent CD27 down-regulation ( p ϭ 0.018) and a larger proportion of CD134 ϩ cells ( p ϭ 0.026) in the memory-derived TCC group. Trends not reaching statistical significance included a larger number of TCC expressing CD62L high in the naive subset-derived group and a greater proportion of CD40L ϩ cells in the memory-derived TCC (data not shown). The percentages of CD2-, CD11a-, CD49d-, CD58-, CD26-, and CD44-positive cells neared 100% in almost all TCC, with comparable fluorescence intensities. Similarly, Ͼ95% of all TCC tested were CD54 ϩ . As expected from in vitro-stimulated T cells, both naive and memory TCC included Ͼ80% of cells with significant HLA-DR surface expression. Although there were only few (Ͻ10%) CD152 ϩ cells in each TCC, variable proportions of CD25-, CD30-, CD80-, and CD86-positive cells were observed, however, without systematic differences. Finally, there were no significant differences in the surface expression of the apoptosismediating molecule CD95 (Fas) on TCC derived from the CD45RA ϩ or CD45RO ϩ subsets. This comprehensive analysis demonstrated that the lower Ag concentration required by CD45RA subset-derived TCC was not secondary to a more favorable expression of molecules for accessory interactions, in particular ligands for costimulation or adhesion molecules. Although it should be pointed out that recognition  Epitope MBP(116 -124). b Epitope MBP(115-125). In patient MS4, TCC reacting to MBP(111-129) were observed from both the naive and memory subset. However, performing a more detailed analysis of these TCC with a set of truncated peptides revealed the recognition of two distinct minimal epitopes, MBP(116 -124) and MBP(115-125), by clones derived from the naive and memory subset, respectively. These clones expressed different TCR (see Results).
c Denotes lack of recognition of the overlapping MBP peptide set despite confirmed specificity to the whole protein.
of epitopes with higher binding affinity for the MHC complex such as MBP(81-99) by naive-derived cells might contribute to their lower EC 50 values, our data show that T cells responding to MBP with overall greater efficiency are selected from the CD45RA ϩ repertoire.

Discussion
In this study, we have dissected the response of highly purified, putative naive and memory-resting CD45RA ϩ /RO Ϫ and CD45RA Ϫ /RO ϩ CD4 ϩ T cell subsets against the human self-Ag MBP. Our findings suggest a major role of the naive T cell subset in the origin and the maintenance of the self-reactive effector CD4 ϩ peripheral T cell repertoire. This notion is first supported by the ϳ2-fold higher frequency of MBP-reactive T cells in the naive CD45RA ϩ /RO Ϫ subset that did not depend on a biased viability, proliferative potential, or coreceptor/adhesion molecule density of the progenitor T cell population. In fact, viability and response to potent stimuli like SAg did not differ in the naive and memory cell subsets. Levels of expression of IL-2R, adhesion molecules, and ligands for costimulation were lower in the purified CD45RA ϩ subset, thus penalizing rather than favoring its proliferation potential (11). Our results are in contrast to recent data reporting a higher frequency of MBP-specific T cells in memory CD45RO ϩ subsets from MS patients (30). The different experimental setting might account for this discrepancy. Burns et al. (30) generated MBP-specific T cell lines from seeding PBMC populations depleted of cells expressing CD45RA or CD45RO. Since not only lymphocytes, but also dendritic cells and monocytes express CD45R isoforms (Refs. 3 and 32 and our observation), isoform depletion may affect Ag presentation by accessory cells, which is crucial for the activation of naive T lymphocytes (6,11,33). Different from that study, we compared the precursor frequency in the two subsets using exact numbers of isolated CD45RA ϩ and CD45RO ϩ T cells and delivering antigenic presentation by means of unmanipulated PBMC.
The second major finding of our study is that the human T cell response to immunodominant MBP epitopes mainly resides in the CD45RA ϩ /RO Ϫ naive subset. This observation is in line with results obtained in experimental models of autoimmunity. Encephalitogenic T cells with specificity for the immunodominant regions of MBP could be selected in vitro from PBL and spleen cells of nonimmunized healthy Lewis rats (34). Similarly, the T cell repertoire of naive healthy rats harbored diabetogenic T cells (35). In addition, CD45RB high cells have autoaggressive potential toward several other target organs (36,37). Unlike the case in inbred animal strains, however, MBP-reactive cells from either the naive or memory CD4 ϩ T subsets of MS patients did not show the strong bias toward a Th1 phenotype which is characteristic for diseasemediating effector cells in experimental models of autoimmune diseases, but were phenotypically more diverse.
Although a prevalent T cell reactivity to MBP from the naive subset in MS patients could also be seen as a lack of involvement of this autoantigen in the pathogenesis of the disease, the detection of proinflammatory and activated MBP-specific T cells in MS patients by different functional assays argues against this possibility (38 -40). Moreover, the increased frequency of MBP-specific T cells which can be raised in the absence of CD28/B7-mediated costimulation in MS patients (41,42) has suggested the presence  of myelin-reactive activated/memory cells. However, MBP-specific responses in those studies were not examined with respect to their origin in the CD45RA ϩ vs CD45RO ϩ compartments. The CD45RA ϩ T cell response to recall Ag increases 3-to 5-fold in the presence of anti-CD28 Ab (43), but it remains to be determined to which extent this population contributes to the overall response of PBMC in the absence of costimulation.
The differences we found in the human CD45RA ϩ and CD45RO ϩ MBP-specific T cell repertoires offer important clues to speculate on the possible mechanisms of regulation and mainte-nance of the autoreactive effector T cell pool. In this context, particularly significant are the more efficient recognition of antigenic epitopes by naive subset-derived TCC and the absence, in each individual subject, of resting memory T cells responding to epitopes recognized by the naive repertoire. Given the lower density of coreceptors like CD28 and CD29 observed on naive subset-derived cells, their Ag-specific responses at 10-fold lower concentrations appear to be significant and indicate that a greater functional avidity of TCR-MHC-peptide interactions outweighed poorer costimulation. The absence within each individual of memory T cells responding to  epitopes recognized by the naive repertoire is difficult to reconcile with the hypothesis that primed MBP-specific cells might revert to a canonical naive phenotype, as shown in other systems (44). Rather, the lack of MBP epitopes shared by the naive and memory subsets in MS patients suggests that naive T cells specific for the immunodominant regions of MBP do not evolve beyond effector stage into resting long-term memory cells with CD45RA Ϫ /RO ϩ phenotype after priming and activation in vivo.
We propose that in MS, upon effective Ag priming probably due to a potent cross-reacting stimulus (45), sufficient TCR affinity, and adequate costimulation, naive CD45RA ϩ T cells become activated and proliferate in secondary lymphoid organs, acquiring a double-positive (CD45RA ϩ /RO ϩ ) effector cell phenotype. These activated cells can reach the brain vascular endothelium via the peripheral blood and cross the blood-brain barrier. Their recognition of immunodominant epitopes of myelin Ag may elicit proinflammatory T cell functions and initiate a demyelinating attack. To limit the autoimmune process and restore self-tolerance, high-affinity MBP-primed T cells need to be neutralized by peripheral mechanisms, including clonal deletion and high-dose anergy (46,47). This concept is supported by the recent observation of a strong T cell response to an immunodominant epitope in MBP-deficient (shiverer) mice, and the inactivation of the high-avidity MBP-specific repertoire in congenic MBP-expressing (C3H) mice (48). According to a recently proposed model (23), memory CD45RO ϩ cells, which we found to recognize subdominant MBP epitopes with low affinity, may have differentiated directly from the naive pool as a result of partial activation, bypassing the proliferative/ effector stage. These cells could survive in the peripheral blood as a result of cross-reactive stimulation, facilitated by their high density of adhesion molecules and receptors for costimulation (46,49).
Our interpretation implies that effector CD4 ϩ cells mediating demyelination in MS are generated de novo from the CD45RA ϩ T cell pool and differentiate either into short-lived effector or longterm memory cells. Defining how naive cells can be recruited into an inflammatory focus leading to organ-specific immune attack warrants further studies. The detection of prevalent, immunodominant autoantigen-specific responses from the CD45RA ϩ /RO Ϫ CD4 ϩ T cell compartment points to this subset as the main reservoir of self-reactive cells, continuously available to trigger human autoimmune processes upon the appropriate stimuli.