Research ArticleMultiple Sclerosis

Immunoglobulin class-switched B cells form an active immune axis between CNS and periphery in multiple sclerosis

See allHide authors and affiliations

Science Translational Medicine  06 Aug 2014:
Vol. 6, Issue 248, pp. 248ra106
DOI: 10.1126/scitranslmed.3008930


In multiple sclerosis (MS), lymphocyte—in particular B cell—transit between the central nervous system (CNS) and periphery may contribute to the maintenance of active disease. Clonally related B cells exist in the cerebrospinal fluid (CSF) and peripheral blood (PB) of MS patients; however, it remains unclear which subpopulations of the highly diverse peripheral B cell compartment share antigen specificity with intrathecal B cell repertoires and whether their antigen stimulation occurs on both sides of the blood-brain barrier. To address these questions, we combined flow cytometric sorting of PB B cell subsets with deep immune repertoire sequencing of CSF and PB B cells. Immunoglobulin (IgM and IgG) heavy chain variable (VH) region repertoires of five PB B cell subsets from MS patients were compared with their CSF Ig-VH transcriptomes. In six of eight patients, we identified peripheral CD27+IgD memory B cells, CD27hiCD38hi plasma cells/plasmablasts, or CD27IgD B cells that had an immune connection to the CNS compartment. Pinpointing Ig class-switched B cells as key component of the immune axis thought to contribute to ongoing MS disease activity strengthens the rationale of current B cell–targeting therapeutic strategies and may lead to more targeted approaches.


Fuelled by recent advances in multiple sclerosis (MS) therapy using CD20-targeted B cell depletion (13), efforts to further understand features of disease-relevant (pathogenic) B cell autoimmunity have increased, both in MS patients and in animal models of MS. Clonal overlap between B cell receptor (BCR) sequences in MS brain parenchyma, meningeal lymphoid follicles, and cerebrospinal fluid (CSF) indicates an immunological continuum inside the central nervous system (CNS) (4, 5). Furthermore, we have previously shown that antigen-driven B cell immune responses inside the CNS compartment are not sequestered from peripheral B cell responses (6), suggesting that disease-driving immune mechanisms against, as yet unknown, identical or highly similar antigenic epitopes operate in the CNS and periphery. Together, these data establish that antigen-experienced B cells provide an immunological connection between numerous different tissues where they may either directly or indirectly support CNS-targeting autoimmunity. The ability to specifically identify and potentially target pathogenic B cells in autoimmune conditions like MS may provide an important basis for the development of immune therapies with improved efficacy and long-term safety. On the basis of previous studies, including our own, it is generally assumed that antigen-experienced B cell subsets provide a pathologically relevant link between the CNS and periphery. However, to date, there is a complete lack of knowledge regarding which subsets of the highly diverse and complex peripheral B cell compartment support CNS-directed autoimmunity.

We applied multicolor flow cytometry sorting of peripheral blood (PB) B cell subsets in combination with next-generation deep immunoglobulin (Ig) heavy chain variable region (Ig-VH) repertoire sequencing (DIRS) of PB B cells and CSF lymphocytes to identify and characterize disease-associated peripheral B cell subsets. We found clusters of clonally related B cells involving CSF-derived Ig-VH and PB CD27IgD B cells, CD27+IgD memory B cells, or CD27hiCD38hi plasmablasts/plasma cells. Our findings strongly suggest that Ig class-switched and/or post–germinal center (GC) B cells provide an immunologically active connection between peripheral and CNS compartments and further support peripheral antigen-driven B cell activation as important contributor to CNS autoimmunity.


Sequencing output

We examined the Ig repertoires in CSF and five different PB B cell subsets (Fig. 1) of eight individual patients with clinically definitive MS (Table 1). Seven of eight patients had more than five oligoclonal bands (OCBs) in their CSF and an increased IgG index (Table 1). All patients were untreated and displayed MS-typical changes on magnetic resonance imaging (MRI) (Table 1). One patient (14711) had active disease at the time of lumbar puncture (LP) (fig. S1); all others had either clinically stable or quiescent relapsing-remitting MS or primary progressive MS without recent contrast enhancement on gadolinium-enhanced MRI (Table 1). The distribution of sorted PB B cells used as input varied across subsets and patients (table S1). CSF IgG-VH sequences were obtained from all patients; CSF IgM-VH transcripts were sequenced for four patients (31012, 34012, 43113, and 43213). From PB of all patients, either IgM-VH or IgM-VH/IgG-VH transcripts were obtained from the following sorted subpopulations: naïve B cells, CD19+IgD+CD27 (IgM-VH); unswitched memory B cells, CD19+IgD+CD27+ (IgM-VH); switched memory B cells, CD19+IgDCD27+ (IgM-VH/IgG-VH); double negative B cells, CD19+IgDCD27 (IgM-VH/IgG-VH); plasmablasts/cells, IgDCD27hiCD38hi (IgM-VH/IgG-VH) (Fig. 1).

Fig. 1. PB B cell subsets connecting to the CNS compartment.

(A) Representative fluorescence-activated cell sorting (FACS) plots (patient 34012) illustrating PB B cell subset identification and sorting strategy; plasmablasts/plasma cells (PC) are CD27hiCD38hi gated on CD19+IgD cells; switched memory (SM), unswitched memory (UM), double negative (DN), and naïve (N) B cells were sorted on the basis of the presence or absence of CD27 or IgD gated on CD19+ cells. Sorting gates are indicated in the FACS plots (see fig. S7 for single-color histograms). PB B cell subsets with connections to the CSF are in black gates; those not connecting to the CSF are in gray gates. (B) PB B cell clusters connecting to the CNS; each column represents the sum of Ig-VH clusters with contributions from CSF and the indicated PB B cell subset.

Table 1. Patient characteristics.

CSF IgG index was calculated using the standard formula [(CSF IgG/CSF albumin)/(serum IgG/serum albumin)]. ID, patient-specific identification number; Dx, diagnosis; RRMS, relapsing-remitting MS; PPMS, primary-progressive MS; EDSS, expanded disability status scale (55) at neurological exam closest to LP; MRI/Gd, presence (+) or absence (−) of gadolinium contrast-enhancing lesions on brain or spinal cord MRI (in parentheses is the time in weeks between last MRI and LP); Tx, therapy at time of sample acquisition; CSF OCB, CSF-restricted oligoclonal bands; CSF WBC, CSF leukocyte count per microliter; CSF volume, CSF volume used to obtain CSF lymphocytes for CSF IgM/IgG-VH DIRS; N/A, not applicable.

View this table:

Each sample was subjected to unbiased 5′-RACE [rapid amplification of complementary DNA (cDNA) ends] and DIRS; separation into IgG-VH– and IgM-VH–expressing B cell subsets was achieved by using isotype-specific (IgM and IgG) reverse primers during Ig-VH 5′-RACE. Because of the generally low numbers of lymphocytes in CSF, we chose to directly perform IgM-VH and IgG-VH 5′-RACE from each CSF cell pellet without sorting of B cell subsets. We obtained a total of ~2.2 million sequencing reads, resulting in ~1.3 million usable IgM-VH or IgG-VH transcripts (table S1). Sequencing reads were deposited at the National Center for Biotechnology Information (NCBI) Sequencing Read Archive (SRA; under accession number SRP042205. Two sequence-clustering approaches were applied for Ig-VH analyses. (i) A distance metric approach [Hamming distance = 1 (7)] was used to group clonally related Ig-VH with highly similar H-CDR3 amino acid sequence, identical H-CDR3 length, and usage of the same IGHV and IGHJ. Clonally related Ig-VH sequences were used to identify and analyze bicompartmental B cell clusters as previously described (6). (ii) Ig-VH data sets of nonredundant Ig-VH reads were generated by considering sequences with identical H-CDR3 and usage of IGHV and IGHJ only once; nonredundant data sets were used to calculate IGHV usage as previously described (6, 8, 9) and to generate somatic hypermutation (SHM) profiles. An overview of the bioinformatics pipeline used for our studies is shown in fig. S2.

Among B cell subsets, the numbers of redundant Ig-VH associated with each nonredundant Ig-VH sequence revealed overall mostly low counts for naïve B cell IgM-VH (N.IgM) and partially very high counts of sequences with identical H-CDR3, IGHV, and IGHJ usage not only in post-GC, Ig class-switched B cells but also in CSF Ig-VH repertoires (fig. S3). Thus, to a reasonable degree, our sequencing approach approximated what is expected biologically: absent clonal expansion among naïve B cells, and extensive clonal activation in B cell subsets resulting from antigen-driven immune responses, such as switched memory B cells and plasmablasts/plasma cells. In addition, this finding supported previous reports of B cell activation in the CNS and CSF (1014).

Here, we used polymerase chain reaction (PCR) amplification of Ig-VH and next-generation immune repertoire sequencing (DIRS). In general, overamplification or sequencing saturation of particularly prevalent Ig-VH transcripts could introduce skewing of data and affect conclusions drawn, particularly with regard to estimating clonal B cell stimulation. Furthermore, in next-generation sequencing, sequencing errors are a potential concern, particularly with respect to DIRS, where read errors could be mistaken for evidence of SHM (15, 16). In the following paragraphs, we will discuss the potential impact of these issues on our results and interpretations where applicable.

Bicompartmental B cell clusters

We identified a total of 46 bicompartmental clusters of clonally related Ig-VH in six of eight patients (14711, n = 15 clusters; 30512, n = 1; 31012, n = 7; 34012, n = 7; 43113, n = 13; 43213, n = 3) (Fig. 1, fig. S4, Table 2, and tables S2 and S3) on the basis of H-CDR3 amino acid sequence similarity and usage of Ig germline segments, features that readily discern Ig-VH sequences of a common origin (17, 18). In all six patients with bicompartmental B cell clusters, we identified CSF IgG-VH clusters connecting to PB IgG+ switched memory B cells (n = 35 clusters). In all four patients from which CSF IgM-VH transcripts were sequenced, we found clusters that were composed of CSF IgM-VH and IgM-expressing switched memory B cells from PB (n = 5 clusters). In three patients, seven CSF IgG-VH clusters connected to PB IgG+ plasma cell clusters, two of which interconnected with IgG+ switched memory B cells (Table 2 and table S2). We also identified a few clusters in which CD19+CD27IgD (double negative) B cells connected to CSF B cells, or in which IgM+ switched memory B cells connected to CSF IgM-VH clusters. A single cluster that included CSF-IgG and both IgG- and IgM-expressing PB B cell subsets was identified in patient 31012 (Table 2 and table S2). No bicompartmental clusters were found in a patient without OCB (29612) and in one other patient, the oldest patient in our cohort (26712, 54-year-old female). It is noteworthy that, for these two patients, no CSF IgM-VH sequences were obtained; it is thus possible that a connection between IgM-expressing PB B cell subsets and the CSF did exist but was not identified for technical reasons. Notably, no connections were identified between CSF and naïve or unswitched memory B cells in PB.

Table 2. B cell clusters connecting the periphery and CSF.

Clusters are grouped by patient; each row represents a bicompartmental cluster. IGHV/IGHJ, closest related variable and joining germline segments used by Ig-VH sequences within each cluster. H-CDR3, representative H chain CDR3 amino acid sequence per cluster; the N-terminal Cys (C), and C-terminal Trp (W) are shown for better visual orientation. Each B cell subset or CSF contributing to bicompartmental clusters is represented in a column showing the number of Ig-VH sequences (#) and the range of SHM along the IGHV sequence portion observed in CSF and PB B cell subsets. Gray shades indicate clusters represented in Fig. 2. SM, switched memory B cells; PC, plasma cells/plasmablasts; DN, double negative B cells.

View this table:

In linear regression analyses, the numbers of bicompartmental IgG-VH clusters obtained did not correlate with CSF white blood cells (WBCs) (r2 = 0.39, P = 0.1) or CSF volume (ranging from 5 to 20 ml) (r2 = 0.12, P = 0.4), suggesting that detection of B cell clusters connecting the CSF and periphery does not depend on either factor (table S4). Similarly, the numbers of bicompartmental IgG-VH were also independent from IgG index (r2 = 0.46, P = 0.07) or total CSF IgG-VH reads (r2 = 0.12, P = 0.4) (table S4). Bicompartmental clusters generally comprised less than 1% of PB B cell subset clusters (fig. S4 and table S5). An exception were bicompartmental clusters involving plasma cells/plasmablasts in patient 14711 comprising nearly 5% of Ig-VH clusters, which may be due to the overall low number of plasma cells/plasmablasts sorted from this patient, but it may also suggest active egress of antibody-producing cells from the CSF (tables S1 and S5).

Affinity maturation in the CNS and periphery

We were interested to determine whether B cells belonging to bicompartmental clusters undergo affinity maturation in the periphery and/or the CNS compartment. Numerous IgM-VH and IgG-VH from either compartment contain multiple SHM along their IGHV-derived portion (Table 2), suggesting that they resulted from antigen-driven immune responses. From our data, it is not possible to draw conclusions whether a B cell response originated in one or the other compartment; however, within clusters of related Ig-VH, we commonly identified a range of SHM (Fig. 2 and Table 2) in the periphery and/or CSF, suggesting that B cells belonging to bicompartmental clusters may have been exposed to antigen stimulation in either compartment.

Fig. 2. Affinity maturation of B cells connecting CSF and PB.

(A to F) Representative lineages of clonally related Ig-VH found in both compartments. Lineages were calculated using IgTree (see Materials and Methods) and visualized in Cytoscape using the proprietary “organic” layout option. Numbers between nodes represent numbers of nucleotide mutations; connections without numbers represent a single-nucleotide mutation. Round nodes without rim encompass a single Ig-VH sequence; round nodes with black rims contain two to nine identical Ig-VH; those with orange rims contain more than nine identical Ig-VH sequences (orange number next to respective nodes). Closest putative germline sequences represent the root of each lineage (black nodes). Gray nodes are lineage intermediates that were not identified as Ig-VH transcript and were calculated by IgTree. Shown also are IGHV and IGHJ used per lineage and representative H-CDR3. Alignments of representative Ig-VH are shown in fig. S5.

Sequencing errors cannot be fully differentiated from true SHM when sequencing natural Ig-VH repertoires where the mRNA templates are not known. In the analyses of our sequencing data, we applied a reductionist approach to only use high-quality Ig-VH sequences for our conclusions regarding SHM occurring in CSF and/or periphery (see Materials and Methods and fig. S2). In 454 sequencing, the probability of sequencing errors due to single and multiple substitutions is less than 10% based on previous quantification (15). Because most nodes in our lineage analysis contain at least two identical in-frame sequencing reads, the probability of one single SHM due to substitution errors is less than 0.01 for a particular node containing two identical sequences [type I error rate less than 1% (15)]; with each additional identical sequencing read contained in a tree node, the probability of sequencing errors is reduced by a factor of 0.1. Accordingly, for three identical sequences to represent a sequencing error, the probability is <0.001, for four sequences <0.0001, for five sequences <0.00001, etc. Thus, lineage relationships of Ig-VH generated by our sequencing approach and calculated using IgTree software (19, 20) support active antigen-driven immune responses to be involved in the shaping of bicompartmental groups of related B cells; representative lineage trees are shown in Fig. 2, and the respective Ig-VH sequence alignments are shown in fig. S5.

IGHV usage in Ig-VH repertoires in PB and CSF

Although our primary focus was to characterize B cells providing an immune axis between CNS and PB, our data also enabled Ig germline gene usage analyses in the overall CSF and PB B cell subset Ig repertoires obtained by DIRS. We found diverse IGHV usage among most PB B cell subsets (Fig. 3 and table S6). Overall, IGHV usage in CSF was limited compared to PB B cells (Fig. 3), whereas IGHV usage by different B cell subsets was generally similar within individual patients. Patient 14711 who had active CNS inflammation with gadolinium contrast-enhancing cervical spinal cord lesions on MRI 2 weeks before LP (Table 1 and fig. S1) displayed more diverse IGHV usage in CSF compared to all other patients (fig. S6). Previous studies, including our own, have shown overrepresentation of certain IGHV4 germline segments in MS CSF (6, 10, 21, 22); in the present study, IGHV4-39 was used by CSF IgG-VH in seven of eight patients, and IGHV4-59/61 by six of eight patients’ CSF IgG-VH (Fig. 3). Possibly artifactual limited IGHV usage was seen in PB B cell subsets for which low cell numbers were obtained by cell sorting, or when fewer nonredundant Ig-VH sequences were available; examples are 26712 double negative IgM, 30512 plasma cell/plasmablast IgM, and 30512 plasma cell/plasmablast IgG (table S1 and Fig. 3).

Fig. 3. IGHV usage in PB B cell subsets and CSF.

Shown are heatmaps representing relative usage of IGHV germline segments by IgM-VH or IgG-VH from the indicated PB B cell subsets or CSF for each patient studied (table S6). To generate heatmaps, IGHV usage values were normalized to the most frequently used IGHV per sample and Ig isotype, which were set as 1. The resulting heatmaps show the most frequently used IGHV in white and the least frequently used IGHV in dark blue (see lower left corner for scale). Notably, IGHV usage is generally limited in CSF (red type), suggesting selective recruitment or survival of B cells in the CSF compartment. IGHV usage of CSF IgG-VH in patient 14711 is more balanced and similar to peripheral B cell repertoires; this patient had signs of active disease 2 weeks before LP (Table 1).

Biased IGHV usage representation could result from overamplification of certain Ig-VH transcripts during PCR or saturation of the sequencing reaction by overrepresented Ig-VH. For our studies, we used 5′-RACE of Ig-VH to eliminate potential biased introduced by IGHV-specific primers and used nonredundant Ig-VH sequence counts to display IGHV usages (see Materials and Methods and fig. S2). We have previously shown similar IGHV usage in PB mononuclear cells (PBMCs) of unrelated individuals (8) using nonredundant Ig-VH data sets. Here, we again found similar IGHV distribution among most PB B cell subsets and in unrelated individuals (Fig. 3), providing indirect evidence that overamplification or sequencing saturation of certain Ig-VH rarely happens. Furthermore, we found low numbers of redundant sequences per nonredundant Ig-VH in naïve B cells, a population where clonal expansion is absent (fig. S3 and table S7).

SHM patterns in Ig-VH repertoires in PB and CSF

Our data were also conducive to understanding the effect of SHM on the IGHV portion of Ig-VH repertoires represented by each PB B cell subset and by CSF IgG-VH and IgM-VH (Fig. 4 and table S8). We were particularly interested in SHM patterns of double negative B cells, which revealed an unexpected immune axis between CSF and PB in our study. As expected, naïve B cells displayed the lowest levels of SHM along their IgM IGHV (Fig. 4A), whereas IgG-expressing switched memory B cells and plasma cells displayed the highest level of SHM (Fig. 4, B and C). SHM profiles of IgM-VH expressed by CD27IgD (double negative) B cells were very similar to those seen in naïve B cells (Fig. 4A). IgG-expressing double negative B cells clustered with IgM-expressing B cell subsets including unswitched memory, switched memory, and plasma cells/plasmablasts (Fig. 4E), overall suggesting lower levels of SHM having shaped the double negative B cell repertoire. Within B cell subsets, IgG-VH had accumulated more SHM compared to IgM-VH (Fig. 4). SHM profiles of CSF IgG-VH appeared most similar to IgG-expressing switched memory and plasma cells/plasmablasts (Fig. 4B), whereas SHM profiles of CSF IgM-VH appeared more similar to IgM-expressing naïve, unswitched memory, and double negative B cell subsets (Fig. 4).

Fig. 4. Dendrogram and heatmaps of PB B cell and CSF Ig-VH SHM profiles.

Shown are heatmaps and hierarchical clustering of Ig-VH SHM profiles; each row represents a PB B cell subset or CSF as indicated by row titles on the right. IgMs are in blue type, and IgGs are in red type; CSF samples are indicated by arrows. Within each subset or CSF sample, SHM profiles were scaled to the highest peak set at 1.0, and colors were assigned such that a count of 0 nonredundant sequences with a certain umber of SHM resulted in blue color and the highest number of sequences with an indicated number of SHM was labeled red; the coloring scale is shown in the upper right. PB B cell subsets or CSF samples clustering together based on SHM profile similarity are indicated by letters on the right (see also Results). (A) Group of subsets characterized by predominantly low SHM, including all naïve B cells and nearly all IgM+ double negative B cells. (B and C) Groups of subsets characterized by extensive SHM and containing all IgG-expressing switched memory B cells and plasma cells, and most CSF samples. (D and E) Subsets characterized by SHM lower than (B) and (C), but higher than (A); these groups include mainly IgM-expressing subsets and IgG-expressing double negative B cells. The data used to generate these heatmaps are shown in table S8.

Overall, the SHM patterns displayed in Fig. 4 support the validity of the B cell subset sorting approach based on CD19, CD27, CD38, and IgD, together with IgM- and IgG-specific PCR amplification of VH. Furthermore, the observed SHM profiles, that is, very low levels of SHM in naïve B cells and the most extensive SHM in IgG-expressing switched memory B cells and antibody-producing plasma cells/plasmablasts (Fig. 4), further suggest that our sequencing approach yielded biologically correct data.


A central goal of MS research is to identify disease-relevant B cells among the vastly diverse peripheral B lymphocyte compartment. We recently demonstrated that an exchange of immunologically active clusters of related B cells occurs between the CNS and PB compartments (6). Here, we describe the identification of specific PB B cell subsets that are clonally related to CSF B cells and could act as important contributors to disease-associated B cell repertoires in MS patients. Overall, our analyses afforded an unprecedented view of Ig germline gene usage and levels of affinity maturation shaping Ig repertoires associated with different B cell subsets in PB and the CSF.

A key question is at which stage during maturation B cells enter the CNS compartment. Influx of naïve B cells would support fully functional secondary lymphoid tissue–like function of the CNS that can operate independent from the periphery. Influx of post-GC B cells, such as memory B cells and plasmablasts/plasma cells, on the other hand, might suggest initial antigen training in the periphery and further affinity maturation and B cell maturation inside the CNS. Overall, our findings suggest that Ig class-switched B cells provide an antigen-experienced immune axis connecting the periphery and the CNS in MS. B cells linking both compartments undergo SHM in PB and CSF, lending strong support (i) to the presence of potentially disease-driving antigens in the periphery and (ii) to the capability of the intrathecal tissues to support active BCR affinity maturation.

The antigens targeted by B cells participating in the immune axis between periphery and CNS remain unknown. It was previously shown that clonally expanded intrathecal plasma cells produce antibodies that bind to as yet unknown myelin or myelin-associated antigens in demyelinating lesions of MS (13), and soluble IgG binding to myelin-oligodendrocyte protein was identified in MS CSF (23, 24). Accordingly, bicompartmental B cell clusters are assumed to participate in potentially CNS-reactive B cell responses. Still, clonal CSF IgG has also been shown to react with Epstein-Barr virus antigens (25), at least raising the possibility that bicompartmental B cell clusters target viral or microbial epitopes and possibly cross-react with CNS epitopes via molecular mimicry, as was shown for T cells in MS (26). Considering that in humans a multitude of viruses are constantly replicating (27), which could stimulate both CNS-specific and nonspecific B cell responses via bystander activation, significant additional work is necessary to understand the involvement of bicompartmental B cell clusters in MS disease activity.

In our study, class-switched memory B cells were found most frequently to provide an immune axis between PB and CSF. Memory B cells are strongly CD20-positive and are efficient antigen-presenting cells (28, 29); memory B cells reactive against myelin have been shown to be present in MS patients’ blood (30). CD20-targeting, B cell–depleting monoclonal antibodies (rituximab and ocrelizumab) are highly effective in reducing or eliminating MS disease activity (13), and in depleting B cells from CSF (31). Our finding that memory B cells participate in an exchange across the blood-brain barrier (BBB) further strengthens the rationale of using anti-CD20–mediated B cell depletion to interrupt disease-promoting peripheral B cell autoimmunity. Natalizumab, a monoclonal anti-VLA4 antibody that also effectively reduces MS disease activity, inhibits transmigration of lymphocytes from PB into the CNS compartment and reduces CSF B cell numbers (32). Two studies described decreased intrathecal production of clonal Ig with natalizumab therapy (33, 34), an effect that could be due to limiting transmigration of B cells involved in OCB production. However, natalizumab was also shown to reduce CSF CD4+ and CD8+ T cells and CD19+ B cells in MS patients (32). Accordingly, intrathecal OCB production may require influx of B cells or T cells to provide signals for B cell differentiation into antibody-expressing plasma cells or plasmablasts, or a combination of both.

Intrathecal plasma cells and B cells are responsible for local production of clonal IgG (35, 36). We found a few instances in three patients of PB plasma cells forming bicompartmental clusters with CSF IgG-VH, which is in agreement with our recent finding that antibodies identical to OCB can theoretically also be produced in the periphery (37). However, it seems unlikely that a sustained supply of clonally restricted peripheral plasma cells entering the CNS compartment is responsible for intrathecal antibody production, because this would require specific recruitment of select plasma cell clones into the CNS. More likely, plasma cells that participate in bicompartmental clusters are progeny of antigen-experienced memory B cells that became established in both the CNS and periphery. An important question that remains is whether the initial antigen exposure of naïve B cells that ultimately mature to bicompartmental memory B cells and OCB-producing plasma cells occurs in the periphery or CNS. Mature naïve B cells that migrate across the BBB could provide input into presumed GC in lymphoid follicle–like structures in the CNS (38, 39). Consistent with this concept, IgD+CD27 naïve B cells have been found in the CSF of MS patients (40) and evidence for B cell differentiation was found in MS CNS (41). In further support of the presence of naïve B cells in CSF, we found CSF IgM-VH harboring very few mutations, suggesting an intrathecal presence of IgM-expressing B cells that have not yet undergone affinity maturation. However, the absence of IGHV usage diversity among CSF IgM-VH in the four patients from which CSF IgM-VH were obtained argues against an ongoing unrestricted exchange of naïve B cells in MS.

We found diverse IGHV usage in patient 14711 CSF, a possible example for a more balanced exchange of B cells during active MS, as previously reported in another MS patient with active disease (6). We also found the largest number of bicompartmental Ig-VH clusters in this patient, but not all IGHV found in this patient’s CSF are also represented in clonally related B cell clusters spanning periphery and CSF. Although speculative, it is possible that the observed clusters represent the pathologically relevant ones participating in active inflammation, whereas B cells expressing IGHV not represented by bicompartmental clusters are pathologically irrelevant. In all other patients’ CSFs, we found limited IGHV usage, suggesting selective survival of B cells in the CNS compartment during disease states that appear quiescent by clinical or MRI measures.

In four of eight patients studied here, we found Ig class-switched double negative B cells that were clonally related to intrathecal Ig repertoires. As previously described by others (42, 43), SHM profiles of double negative IgG-VH are frequently shifted toward fewer somatic mutations when compared to IgG+ switched memory B cells, which may support double negative cells as progenitors of switched memory B cells. However, on the basis of a high-throughput immune repertoire sequencing study comparing Ig-VH of switched memory and double negative B cells, Ig class-switched CD27 and CD27+ B cells may also be developmentally related in either direction, that is, with double negative B cells being precursors of switched memory B cells and vice versa (44). PB double negative B cells are expanded in patients with active systemic lupus erythematosus; these cells have been proposed to represent a class of memory B cells that fail to undergo productive GC maturation (42). B cells accumulating SHM outside of GCs likely also escape the scrutiny of peripheral B cell tolerance and may harbor increased autoimmune potential (45). Our observation in this cohort of MS patients, that SHM profiles of IgM-expressing double negative B cells compare more closely to naïve B cells than unswitched memory B cells, suggests that they are direct progeny of naïve B cells and undergo class switch recombination (CSR) outside of GCs or exit GCs prematurely. Double negative B cells proliferate more readily upon Toll-like receptor–mediated activation than with BCR cross-linking, suggesting that they function in an antigen-independent manner (42). Further work will be necessary to understand the relevance of double negative B cells to MS; however, our findings suggest that double negative B cells may be a relevant part of the immune axis linking the periphery with the CNS.

The identification of PB B cell subsets participating in bicompartmental immune response would not have been possible without high-throughput next-generation sequencing techniques that are far more comprehensive than Sanger sequencing–based approaches. Nonetheless, there are several limitations of our approach that should be recognized.

(i) We are unable to directly measure clonal expansion. However, the number of redundant Ig-VH associated with nonredundant reads appears to approximate biologically expected findings. With the technology used here, it is difficult to determine whether a given sequence appears more frequently because of a large quantity of its mRNA being recovered from a single cell or multiple cells expressing that same sequence, or because of disproportional PCR amplification or sequencing of certain Ig-VH transcripts. Higher number of reads per nonredundant Ig-VH could also reflect low sequence diversity within these populations and thus sequencing oversaturation. New technologies incorporating nucleotide tags (46, 47) or high-throughput single-cell sequencing technologies (48) may in the future permit better estimates of the numbers of mRNA molecules obtained from single cells, thereby permitting more accurate measurements.

(ii) Both identifying clonal overlaps and appropriate representation of IGHV usage depend on the sampling and sequencing depth of the interrogated compartments, and the likelihood of detecting clonally related Ig-VH between CSF B cells and PB B cell subsets is expected to correlate with the level of clonal expansion in each subset. Clonal expansion is a characteristic of activated memory B cells or plasma cells, theoretically facilitating identification of B cell clusters by DIRS. Conversely, absent or low-level clonal expansion [that is, homeostatic proliferation (49)] found in the naïve B cell compartment likely impedes detection of PB naïve B cells related to CSF Ig-VH, and greater numbers of cells sampled and increased sequencing depth may, or may not, reveal additional clusters.

(iii) Sequencing errors introduced by 454 sequencing could result in artifactual clonal families of Ig-VH sequences (16). In our experiments, we studied natural Ig-VH repertoires without knowledge of template sequences. It is thus impossible for us to determine exactly which mutations are sequencing errors and which result from SHM. We applied a reductionist approach to select Ig-VH sequences for lineage analyses that incorporate likely true SHM, rather than sequencing errors. SHM profiles resulting from PB B cell subsets reveal biologically expected information (that is, low SHM in naïve B cells and extensive SHM in IgG-expressing memory B cells and plasma cells) and therefore suggest that, overall, sequencing errors do not exert extensive influence on the representation of SHM.

Despite these limitations, our data in summary reveal a highly dynamic and interconnected B cell repertoire shaping the PB and CNS immune landscape in MS. Our study represents an important initial snapshot of processes that are likely to come into better resolution as additional observations—especially those accompanying the early phases of disease development, during clinical attacks, and with transition to the late progressive phase of MS—are revealed. Indirectly, our findings suggest that exchange of disease-associated B cell subsets between the CNS and periphery in MS is a highly selective process. This process may be supported by B cells that have undergone CSR and/or affinity maturation, in the periphery and/or CNS compartment. Although the target antigens and exact mechanisms driving the establishment of intrathecal B cell repertoires remain unknown, the ability to identify antigen-experienced B cells participating in MS immune pathology, as described here, should help to resolve these key issues. In particular, identifying the specificity of bicompartmental B cells will ultimately link B cell exchange between periphery and CNS to MS immune pathology.


Patient samples

We studied eight patients with clinically definitive MS (Table 1) based on the latest diagnostic criteria for MS (50). Patients were selected on the basis of availability of paired CSF and sufficient PB. On the basis of a previous study (6), we expected to identify bicompartmental clusters in about 80% of patients. No sample size calculations were performed before our experiments; data analyses were not performed blinded because no comparisons between different groups of patients were performed. PB (48 ml) was obtained via standard venipuncture; 5 to 20 ml of CSF were collected during standard LP during the same visit. To avoid potential CSF contamination with PB, the first 2 ml of CSF was discarded. Immediately after LP, the entire volumes of CSFs were centrifuged at 400g for 15 min, and cell pellets were lysed in RLT buffer (Qiagen RNeasy kit). PBMCs were isolated using a Ficoll gradient, red blood cells were lysed, and PBMCs were washed in phosphate-buffered saline (PBS) containing 1% bovine serum albumin (BSA). These studies were approved by the institutional review board of the University of California, San Francisco, and informed consent was obtained from patients before CSF/PB collection.

Cell staining and sorting

After FcR blocking (mouse serum; Jackson ImmunoResearch), PBMCs were incubated for 20 min on ice in the dark with the following antibodies: CD19 [allophycocyanin (APC) Cy7], IgD [phycoerythrin (PE) Cy7], CD27 (Qdot605), CD38 [peridinin chlorophyll protein (PerCP) Cy5.5], and CD3 (Pacific blue). Then, cells were washed and resuspended in PBS containing 1% BSA; 4′,6-diamidino-2-phenylindole (DAPI) was added to discriminate dead cells. B cell subsets were identified using the expression of the following surface markers and collected on a FACSAria (BD Biosciences): naïve, CD19+IgD+CD27; unswitched memory, CD19+IgD+CD27+; switched memory, CD19+IgDCD27+; and double negative, CD19+IgDCD27. Plasmablasts/cells were gated on IgD cells and selected on the basis of high expression of CD27 and CD38 (IgDCD27hiCD38hi) (Fig. 1). Gating controls (single-color histograms) for CD19, CD20, CD27, CD38, and IgD are shown in fig. S7. Dead cells and T cells were gated out using DAPI and CD3 expression, respectively. B cell subsets were sorted and immediately lysed in RLT buffer (Qiagen RNeasy kit) and stored at −80°C.

Unbiased Ig mRNA amplification and Ig repertoire sequencing

Sequencing work flow was performed as previously described (8). In brief, total RNA was isolated from CSF (miRNeasy Mini Kit, Qiagen) and PB B cell subsets (RNeasy micro kit, Qiagen). RNA quality was assessed using an Agilent Bioanalyzer. Total isolated RNA was reverse-transcribed (SMARTer RACE, Clontech), and about 27% of each cDNA reaction was used for IgM-VH and IgG-VH amplification via PCR using SMARTer RACE 10× Universal Primer Mix (Clontech) and IgG or IgM isotype-specific 3′ primers (IgG, 5′-GGGAAGACSGATGGGCCCTTGGTGG-3′; IgM, 5′-GATGGAGTCGGGAAGGAAGTCCTGTGCGAG-3′) for 31 cycles following the manufacturer’s recommendation. These primer sequences were attached to Lib-L–specific adaptor (454 sequencing, Roche) and barcode sequences to amplify IgG-VH and IgM-VH libraries. Barcoded IgM-VH [~715 base pairs (bp)] and IgG-VH (~640 bp) transcript libraries were purified using AMPure XP (Beckman Coulter Inc.), quantified using PicoGreen (Life Technologies), and normalized to 1 × 109 molecules/μl. Sequencing library pools of uniquely barcoded B cell subsets from each subject were created by pooling the normalized IgM-VH and IgG-VH amplified libraries at relative proportions to obtain 1× (double negative and plasma cells/plasmablasts), 2× (naïve), and 4× (switched memory and unswitched memory) sequencing depth, respectively. The final multiplexed library pools were subjected to emulsion PCR and unidirectional sequencing using the GS FLX Titanium Lib-L chemistry (454 Sequencing, Roche).

Sequence analysis

Figure S2 illustrates and summarizes the bioinformatics pipeline used for Ig-VH sequence analysis. For all reads, IGHV and IGHJ gene segment usage, CDR1–3 amino acid sequence, and number of SHM events were determined using VDJFasta as previously described (9). In general, Ig-VH sequences were considered related on the basis of the fact that IGHV and IGHJ usage are stable features of clonally related B cells, that H-CDR3 sequence is identical or highly similar between clonally related B cell, and that the H-CDR3 sequence length is virtually invariable within groups of clonally related B cells. For calculations of IGHV usage frequency, nonredundant Ig-VH sequence counts were considered; that is, sequences with identical H-CDR3, IGHV, and IGHJ were counted only once. The numbers of nonredundant IgG/M-VH are shown in table S1; per IGHV germline segment usage is displayed as percentage of all Ig-VH sequences obtained per sample. To generate SHM profiles, the most frequent SHM count encountered in a group of nonredundant Ig-VH was used. This approach ensured correction for potential overamplification of more abundant Ig-VH mRNAs.

To identify bicompartmental clusters containing IgM-VH or IgG-VH from CSF and PB B cell subsets, clonally related Ig-VH sequences were grouped on the basis of H-CDR3 similarity and usage of identical Ig germline segments. This approach combined Ig-VH containing identical H-CDR3 amino acid sequences and Ig-VH with H-CDR3 differing by a single amino acid compared to all other H-CDR3 in a cluster based on a Hamming distance = 1 (6, 7). For lineage analysis, only Ig-VH sequences for which in-frame H-CDR3 were identified and that spanned at least from the 5′ end of H-CDR1 to the 3′ end of H-CDR3 with a contiguous reading frame were considered; all out-of-frame sequences were thus automatically filtered on the basis of the presence of uncommonly occurring insertions and deletions (indels). Putative germline sequences were obtained using SoDA (34) and were used as tree root nodes (black) in our lineage tree calculations.

We used IgTree (19, 20) (provided by R. Mehr, Bar-Ilan University, Ramat-Gan, Israel) to calculate B cell lineages contained within each IgG-VH cluster. This software has been successfully used with DIRS data (20). Automated multiple alignments of cluster-derived Ig-VH sequences and the corresponding germline sequence were performed using ClustalW 2.1 (51). Lineage trees were displayed in Cytoscape [version 2.8.3 (52)] using the proprietary organic layout, which permits a more compact depiction of lineages; tree nodes were colored according to their origin (CSF and PB cell subsets) and isotype (IgM and IgG). Putative germline nodes are labeled black, and lineage intermediates not found by DIRS were calculated by IgTree and are labeled gray. The size of tree nodes is proportional to the number of identical sequences contained within each node. Ig-VH amino acid alignments were generated using the MAFFT algorithm (53) within Jalview (54) software.


Linear regression analyses and analysis of variance (ANOVA) were performed in GraphPad Prism 6.0. Comparisons of numbers of reads in each nonredundant Ig-VH sequence group between naïve PB B cells and all other B cell subsets and CSF per patient (table S7) were performed using Kruskal-Wallis test (ANOVA with Dunn’s multiple comparisons) (GraphPad Prism 6.0).


Fig. S1. Sagittal cervical spinal cord MRI of patient 14711.

Fig. S2. Bioinformatics pipeline for Ig-VH analyses.

Fig. S3. Numbers of Ig-VH in CSF and B cell subsets approximate expected levels of antigen stimulation.

Fig. S4. Bicompartmental clusters relative to total PB B cell clusters.

Fig. S5. Ig-VH amino acid sequence alignments of bicompartmental clusters shown in Fig. 2 (separate PDF file).

Fig. S6. IGHV usage profiles of CSF Ig-VH.

Fig. S7. Gating controls (single-color histograms) for CD19, CD20, CD27, CD38, and IgD flow cytometry.

Table S1. Sorted cells and sequence counts.

Table S2. PB B cell clusters connecting to the CSF/CNS compartment.

Table S3. Number of bicompartmental clusters by patient and B cell subset expressing Ig-VH related to CSF Ig-VH.

Table S4. Correlations between number of bicompartmental clusters found and CSF metrics.

Table S5. Bicompartmental clusters in proportion to total PB B cell clusters of the indicated PB B cell subset.

Table S6. IGHV usage data used to generate heatmaps displayed in Fig. 3 (separate Excel file).

Table S7. Comparisons of numbers of reads in each nonredundant Ig-VH sequence group between naïve PB B cells and all other B cell subsets and CSF per patient.

Table S8. SHM data used to generate heatmaps displayed in Fig. 4 (separate Excel file).


  1. Acknowledgments: We are deeply grateful to our patients who have agreed to participate in this research study. We thank R. Mehr (Bar-Ilan University, Ramat-Gan, Israel) for providing IgTree. We thank S. Potluri and H. Wan (both Rinat-Pfizer) for helpful discussions and suggestions. Funding: These studies were supported by grants from the National Multiple Sclerosis Society (RG-4868 to H.-C.v.B.), the NIH/National Institute of Neurological Disorders and Stroke (K02NS072288 to H.-C.v.B.), and Pfizer Inc. H.-C.v.B. was also supported by an endowment from the Rachleff Family Foundation. Author contributions: A.P. performed experiments, analyzed and interpreted data, and wrote the paper; L.A. performed bioinformatics analyses, interpreted data, and wrote the paper; T.C.K. performed experiments, analyzed and interpreted data, and wrote the paper; M.S. performed bioinformatics analyses, interpreted data, and wrote the paper; S.W. performed bioinformatics and statistical analyses and wrote the paper; S.J.P. performed bioinformatics analyses and wrote the paper; P.D.S. performed experiments and wrote the paper; D.T. performed experiments; L.Z.Z. performed experiments; M.D. collected patient samples and provided clinical information; A.A. collected patient samples and provided clinical information; S.L.H. interpreted data and wrote the paper; H.-C.v.B. planned all experiments, analyzed data, and wrote the paper. Competing interests: A.P. is a current employee of Biogen Idec, Cambridge, MA. S.L.H. has a paid consulting relationship with BioMarin. H.-C.v.B. has received research funding from Pfizer Inc. and F. Hoffmann-La Roche, and consulting fees from Novartis. The other authors declare that they have no competing interests. Data and materials availability: Sequencing reads were deposited at the NCBI SRA ( under accession number SRP042205.
View Abstract

Stay Connected to Science Translational Medicine

Navigate This Article