Peripheral B‐cell dysregulation is associated with relapse after long‐term quiescence in patients with multiple sclerosis

Abstract B cells play a major role in multiple sclerosis (MS), with many successful therapeutics capable of removing them from circulation. One such therapy, alemtuzumab, is thought to reset the immune system without the need for ongoing therapy in a proportion of patients. The exact cells contributing to disease pathogenesis and quiescence remain to be identified. We utilized mass cytometry to analyze B cells from the blood of patients with relapse‐remitting MS (RRMS) before and after alemtuzumab treatment, and during relapse. A complementary RRMS cohort was analyzed by single‐cell RNA sequencing. The R package “Spectre” was used to analyze these data, incorporating FlowSOM clustering, sparse partial least squares‐discriminant analysis and permutational multivariate analysis of variance. Immunoglobulin (Ig)A+ and IgG1 + B‐cell numbers were altered, including higher IgG1 + B cells during relapse. B‐cell linker protein (BLNK), CD40 and CD210 expression by B cells was lower in patients with RRMS compared with non‐MS controls, with similar results at the transcriptomic level. Finally, alemtuzumab restored BLNK, CD40 and CD210 expression by IgA+ and IgG1 + B cells, which was altered again during relapse. These data suggest that impairment of IgA+ and IgG1 + B cells may contribute to MS pathogenesis, which can be restored by alemtuzumab.


INTRODUCTION
Alemtuzumab is a humanized monoclonal antibody targeting CD52-expressing cells, which are primarily expressed by T and B cells. Alemtuzumab, which is used to treat people with multiple sclerosis (MS), leads to the depletion of T-cell subsets as well as unswitched memory, switched memory and double-negative B cells, with an overall increase in total B cells accompanied by a relative increase in the proportion of transitional and na€ ıve B-cell subsets during repopulation. [1][2][3][4][5][6][7] This dramatic change in B-cell subsets is hypothesized to explain the success of alemtuzumab as an MS disease-modifying therapy (DMT). 8 T-cell repopulation is much slower and can take years. 5,7 A remarkable feature of this DMT is that patients with MS treated with alemtuzumab can display long-lasting protection from disease progression that lasts years without the need for continuous treatment. 9 However, a subset of patients with MS treated with alemtuzumab undergo inexplicable relapse after the second course. Understanding the reasons for this relapse will allow us to better predict treatment failure and identify the specific immune cells responsible.
One explanation for the relapse occurring in some patients with MS treated with alemtuzumab is a defect in the repopulating B-cell compartment. To investigate this, we have used mass cytometry combined with FlowSOM clustering 10 and sparse partial least squares-discriminant analysis (sPLS-DA) to identify and analyze 33 distinct Bcell subsets in patients with MS before and after alemtuzumab treatment, including long-term stable patients with MS as well as those who have relapsed.
To determine whether the overall B-cell repertoire differed between groups (and to identify metaclusters that contributed to these differences), an sPLS-DA was performed. The quantity of each metacluster was calculated as a proportion of B cells and the absolute cell count for each patient. The sPLS-DA plots suggested there was a difference between groups (Figure 2a, d, g), which was confirmed by a permutational multivariate analysis of variance (PERMANOVA) using the selected metaclusters highlighted by the sPLS-DA. Of particular note, there was a clear difference between untreated patients/controls compared with patients with MS treated with alemtuzumab. Statistically significant differences were found between non-MS and prior, non-MS and post-2, non-MS and relapse, prior and post-1, prior and post-2, prior and relapse (Figure 2b, e, h). The metaclusters used in the first two components are shown (Figure 2c, f, i). Each row represents a different metacluster, with annotations indicating the conventional B-cell subset each metacluster belongs to. Metaclusters are ranked based on their contribution to each component, with the largest contributor at the bottom. In combined proportion + count (Figure 2a-c), both proportion and count data from multiple B-cell subsets contributed to differences between the groups. This included contributions from subsets previously reported to be involved in MS pathogenesis and affected by alemtuzumab treatment (transitional, na€ ıve, doublenegative, unswitched memory and switched memory B cells) as well as novel subsets, particularly IgA + CD20 + , IgG  Figure 1. Clustered B-cell repertoire differs between disease and treatment groups. FlowSOM clustering was first done on B cells from each patient. A total of 40 metaclusters were generated. (a-c) Data were then downsampled evenly across groups to 6000 cells to create FIt-SNE plots. Colors represent labeled parameter (density, metacluster, CD20, CD24, CD27, CD38, IgA, IgD, IgG 1 , IgG 2 , IgG 3 ). (d) A heatmap was generated to visualize the relative marker expression (columns) across metaclusters (rows). Each marker was rescaled by minimum/maximum, to compare high (yellow) and low (black) marker expression. Median signal intensity for each marker was calculated for each metacluster. Markers (columns) were rescaled by minimum/maximum. Each metacluster was assigned to one of the conventional B-cell subsets (transitional, na€ ıve, CD24 hi na€ ıve, unswitched memory, switched memory, double negative, IgG 3 , IgG 1 , IgG 2 , IgA + CD20 + , IgA + CD20 À , CD20 À B cells). FIt-SNE, fast interpolation-based t-distributed stochastic neighbor embedding; Ig, immunoglobulin. n). FIt-SNE plots visualize differences between groups ( Figure 3e). IgA + CD20 + CD27 À CD268 + B cells were significantly higher in patients with MS prior to alemtuzumab compared with non-MS controls (Figure 3f

BLNK and CD210 expression on B cells are significantly lower in patients with MS
We made use of our comprehensive mass cytometry panel to interrogate B-cell expression of a range of markers that may be associated with MS pathogenesis and/or affected by alemtuzumab treatment (Supplementary figures 4 and 5). We discovered that although BLNK, CD40 and CD210 were constitutively expressed across most subsets (Figure 4a), the overall expression of these markers by B cells was lower in patients with untreated MS compared with non-MS     IgA-155Gd

IgG2-209Bi
CD11c-172Yb    (Figure 4b). To confirm our observation in a second cohort and determine whether similar changes occurred at the transcriptomic level, single-cell and total RNA sequencing data were analyzed on the publicly available data set GSE133028. 11 Conventional B-cell subsets constitutively expressed BLNK, CD40 and IL10RA (CD210) with changes in expression as the subsets become more developed (Figure 4c), mirroring those which we observed in our cohort (Figure 4a). For singlecell data, B cells were identified by FlowSOM clustering (Figure 4d), and transcript levels were compared between healthy controls and patients with untreated relapsingremitting MS (RRMS; Figure 4e). As seen in our mass cytometry cohort (Figure 4b), there was a downward trend of BLNK, CD40 and IL10RA (CD210) in patients with RRMS (n = 12) compared with healthy controls (n = 3). Finally, we confirmed that BLNK and CD210 expressions were significantly lower in our newly identified subsets of interest: IgA + CD20 + (Figure 5a-c), IgG 1 + (Figure 5d, e) and IgG 2 + (Figure 5f, g) B cells in people with MS. After treatment with alemtuzumab, BLNK and CD210 expressions were restored to healthy control levels in IgA + CD20 + CD27 + CD268 + B cells (Figure 5b), IgG 1 + CD27 À CD80 low (Figure 5d) and both IgG 2 + B-cell subsets (Figure 5f, g). In patients that relapsed, BLNK expression in these subsets was again lower compared with post-2. CD40 expression was lower in patients with relapse compared with non-MS controls. Together, these results suggest a functional impairment of these B-cell subsets may contribute to disease pathogenesis, which can be restored by alemtuzumab treatment.

DISCUSSION
Recent advances in MS therapeutics suggest that B cells play a prominent role in the pathogenesis of MS, but how targeting B cells suppresses disease activity is not clear. Key findings to emerge from our analyses were the unexpected changes occurring in circulating IgA + , IgG1 + and IgG2 + B cells. Using the power of mass cytometry, we were able to identify seven discrete B-cell subsets that were affected by MS disease and altered by alemtuzumab therapy based on their surface expression of CD27, CD80, CD184 and CD268. Deeper interrogation of these novel B-cell subsets revealed changes in the expression of BLNK, CD40 and CD210 during MS pathogenesis and following the administration of a long-lasting B-cell-targeting DMT. B-cell-derived Igs, particularly IgG 1 , have long been associated with autoimmunity, 12,13 including MS. 14,15 Serum levels of IgG 1 are elevated in patients with MS compared with healthy controls, 16 with alemtuzumab decreasing total serum IgG in patients with MS. 17 While we did not detect any difference in circulating IgG 1 + Bcell subsets between non-MS controls and untreated patients with MS, we did observe a significant DMT effect with significantly lower numbers of IgG 1 + B cells following alemtuzumab. Suggestive of a pathogenic role for IgG 1 + B cells was the significant rise we observed for this subset in the three patients who relapsed. Serum IgG 2 levels in patients with MS have been found to be either no different compared with healthy controls 16 or decreased in both patients with MS and clinically isolated syndrome compared with healthy controls. 18 This latter study is consistent with the changes in and patients with MS post-1 (up to 12 months after alemtuzumab dose, n = 9), post-2 (greater than 12 months, n = 10) alemtuzumab and relapse (n = 3)], a PERMANOVA was performed followed by pairwise comparisons with Holm's correction. Prior, post-2 and relapse groups were compared with non-MS controls (for three comparisons). A LMM was calculated when comparing between patients with MS before and after treatment. A total of 4999 permutations were then run to calculate P-values. Five multiple comparisons were made (prior to post-1, post-2 and relapse; and post-1 to post-2 and post-2 to relapse) using a further 4999 permutations with Holm's correction. The mean is shown in non-MS controls, P-values < 0.1 are shown. (c) Five B-cell subsets, na€ ıve (IgD + CD27 À ), double negative (IgD CD27 À ), unswitched memory (IgD + CD27 + ), switched memory (IgD À CD27 + ) and plasmablasts/plasma cells (IgD À CD27 hi ), were first sorted and then underwent total RNA sequencing and calculated RPKM. The levels of BLNK, CD40 and IL10RA (CD210) were calculated. Solid lines signify data are available for adjacent timepoints. A LMM was calculated when comparing between patients with MS before and after treatment. A total of 4999 permutations were then run to calculate P-values. Each group was compared with each other (10 comparisons) using a further 4999 permutations with Holm's correction. The mean is shown for each subset. (d) Single-cell RNA sequencing was performed on PBMCs from patients with RRMS and healthy controls. Clustering was first done to identify B cells, with FIt-SNE plots for visualization. (e) Transcript levels of BLNK, CD40 and IL10RA. Permutation of t-test, the mean is shown. 24 N, CD24 hi B cells; A20 + , IgA + CD20 + B cells; A20 À , IgA + CD20 À B cells; CD20 À , CD20 3 IgG 2 + CD27 À CD184 + B cells we observed following treatment and during relapse. The dichotomy between IgG 1 and IgG 2 is well known, as cytokines such as interferon-c can inhibit IgG 1 while promoting IgG 2 secretion 19 and may contribute to the differences between IgG 1 + and IgG 2 + B-cell levels after DMT and during relapse. The role of IgA-producing B cells in MS pathogenesis is coming under increased scrutiny. In preclinical models, IgA + B cells have been found to play a protective role against experimental autoimmune encephalomyelitis. 20 Similarly in humans, IgA + CD19 + CD27 + B cells produce more interleukin (IL)-10 compared with IgD + CD27 À , IgM + CD27 + and IgG + CD27 + B cells when stimulated with IL-21 and CD40L. 21 Therefore, the higher number of IgA + CD20 + B cells in patients with MS prior to treatment with alemtuzumab compared with non-MS controls may reflect an immunological rheostat attempting to contain the autoinflammation underway. Similar results were observed in a recent study. 22 This hypothesis is supported by the fact that these B-cell subsets increase upon relapse.
BLNK was first identified as a central linker protein that connects the B-cell receptor-associated kinases with other signaling pathways. 23 BLNK plays an important role in Bcell development. 24,25 We have discovered that compared with non-MS controls, BLNK protein levels in people with MS are significantly lower in all circulating B cells analyzed, particularly the IgA + , IgG1 + and IgG2 + subsets we have identified to play a novel role in MS pathogenesis and their response to alemtuzumab. Our observation is supported by a recent study showing lower BLNK mRNA levels in patients with benign MS compared with healthy controls. 26 Lower BLNK levels is likely to result in perturbed B-cell receptor signaling 27 and/or B-cell survival 28 in MS. It may also indicate a defect in immune regulation as mice deficient in BLNK have impaired IL-10 production from B10 cells and develop more severe experimental autoimmune encephalomyelitis. 29 These differences in BLNK expression do not appear to be an inherent defect in the B-cell compartment, as B cells emerging from the bone marrow following the first round of alemtuzumab treatment had BLNK levels that were comparable to non-MS controls. However, lower BLNK expression in B cells does appear to be a harbinger of disease progression and a potential early indicator of MS relapse.
Like BLNK, levels of the IL-10 receptor (CD210) were lower in almost all B-cell subsets analyzed from patients with MS compared with non-MS controls. This suggests that circulating B cells in patients with MS are likely to be resistant to the effects of IL-10 on growth, differentiation 30 and survival. 31 CD210 expression has also been associated with regulatory B-cell activation in both mice 32 and humans. 33 Together with signaling through CD40, B cells respond to IL-10 by altering their production of IgM, IgG 1 and IgG 3. 34 Indeed, most B cells, including some of those affected by MS disease and altered by alemtuzumab therapy, expressed lower levels of CD40, a well-known MS risk gene. 35 Signaling through CD40 and IL-21 receptors drive B10-cell development and expansion in mice, which can inhibit established experimental autoimmune encephalomyelitis. 36 In humans, CD40 activation can induce B-cell production of IL-10 and upregulation of CD210. 33,37 B cells from patients with MS produce less CD40-induced IL-10. 38,39 A limitation within the study was the low number of relapse patients (n = 3). Although there were significant differences in B-cell subset levels between groups, more may be observed with a larger patient cohort. Furthermore, full kinetics from individual patients from prior to relapse were absent because of limits with sample collection. Samples from patients over the full period would be beneficial in future studies. All three relapse patients had also previously received other DMTs prior to alemtuzumab, with no samples taken prior, so the effects observed may not be fully dependent on alemtuzumab. Although B cells undoubtedly play a major role in the pathogenesis of MS, other immune cells (such as T cells 5 ) would be of interest in future studies.
Our in-depth immunophenotyping of B cells has provided new insights into the effect of alemtuzumab on B-cell subsets. The lower level of BLNK, CD40 and CD210 expression in patients with untreated MS likely influences the function of various B-cell subsets, including IgA + CD20 + and IgG 2 + B cells in controlling disease. The increase in IgG 1 + B cells during relapse likely contributes to disease pathogenesis, possibly because of a breakdown in regulatory mechanisms. The data shown here highlight the complexity of MS, and the contribution of many B-cell subsets that will influence disease outcome. Future therapeutics may benefit from restoring BLNK and/or CD210 levels in an effort to inhibit the activation of IgG 1 + B cells while promoting the regulatory functions of IgA + B cells.

Study participants
Ethical consent for the study was obtained from the Research Integrity and Ethics Administration of the University of Sydney (project numbers 2018/708 and 2018/377). MS was defined by McDonald 2017 criteria and disease activity was defined by neurological signs/symptoms and the presence of new T2 or gadolinium-enhancing magnetic resonance imaging lesions. No patients were on DMTs at the time of blood sampling. All patients had low Expanded Disability Status Scale scores (range 0-2.5). Patient data are shown in Table 1.
A total of 23 patients with MS were included in this study. All patients were reviewed clinically every 6 months and had 3-T magnetic resonance imaging of brain and spinal cord prior to and 6 months after starting treatment and then every 12 months unless new symptoms developed. Blood samples taken from patients with MS prior to alemtuzumab treatment   (prior, n = 11) were treatment na€ ıve (5/11) or free from treatment for at least 1 month prior to alemtuzumab (6/11). The first course of alemtuzumab was given for 5 consecutive days, with a repeated second course 12 months later over 3 days. Post-alemtuzumab treatment was divided into three groups. Post-1 had blood taken up to 12 months after alemtuzumab dose (8-11 months after the first dose n = 3; 6-12 months after the second dose n = 6). One of these patients did not have the peripheral blood-derived mononuclear cells (PBMCs) counted, so eight of nine patients were included for cell count data. Post-2 had blood taken 29-39 months after first treatment (> 12 months since second course; n = 10).
Relapse included patients that developed new clinical symptoms after two courses of alemtuzumab (with new lesions confirmed by magnetic resonance imaging) and blood was taken within 1 week of symptoms and prior to corticosteroid or additional alemtuzumab treatment (n = 3). Age-and sexmatched non-MS controls (n = 9) were included in the study, with samples taken from a single timepoint (Table 1).

Blood sampling
PBMCs were isolated from blood within 0-8 h of collection in EDTA vacuette tubes (Greiner Bio-One International, Kremsm€ unster, Austria) using a Ficoll-Paque PLUS (GE Healthcare, Chicago, IL) density separation gradient. When blood was not processed immediately, samples were stored at room temperature to minimize cell loss. 40 Although leaving cells at room temperature can alter receptor expression (particularly chemokine receptors), 40 variance of receptor expression between patients was comparable to internal controls, suggesting that experiment/instrument variability played a larger role than blood processing time (Supplementary figure 5). Cells were counted and blood volume was recorded, such that the concentration of cells in blood could be calculated. Samples were cryopreserved in 5% dimethyl sulfoxide/fetal bovine serum for storage in liquid nitrogen prior to mass cytometry staining.
Cell staining and analysis by mass cytometry About 2.5 9 10 6 cells were resuscitated by thawing in a 37°C water bath and washed in Roswell Park Memorial Institute medium. Individual patient/timepoint samples were first barcoded with anti-human CD45 (on four different metal isotopes) and purified human FcR-binding inhibitor (eBioscience Inc., San Diego, CA) for 30 min, such that four independent samples could be combined for further staining as described. 41 To control for batch variability, PBMCs taken from a single non-MS control (taken from a single timepoint) were included within each batch as an internal control, used across 25 batches (Supplementary figure 5). These controls were analyzed in the same manner to allow comparisons between batches. Samples were combined (for 7.5-10 9 10 6 cells) and stained with cisplatin (Fluidigm, South San Francisco, CA) for 5 min as a live/dead marker. Cells were stained with antibodies specific for the markers in Supplementary table 1

Mass cytometry data analysis
Samples were initially gated as shown in Supplementary  figure 1a, b using FlowJo version 10.4 (Becton Dickinson, Ashland, OR). All analyses were performed on single live CD3 À CD19 + CD20 +/À B cells.
Heatmaps were generated using pheatmap 45 as part of the Spectre package 43 in R. Each marker (column) was rescaled by minimum/maximum for each metacluster (row).
Manual gating was used to confirm changes identified by clustering results. 46 B cells were differentiated into conventional B-cell subsets and further defined using markers in Supplementary table 1. The quantity of each subset was calculated as a proportion of total B cells. When patient cell counts were made available, the proportion data were used to calculate the absolute number of each B-cell subset.

Mass cytometry statistical analysis
All statistics were calculated using packages available within R, 44 using Type III Sum of Squares for PERMANOVA. sPLS-DA was performed using the "mixOmics" package in R 47 for feature selection of metaclusters that differ between groups of interest. A principal component analysis reduces the number of dimensions by summarizing the overall variance of a data set. The group that each individual belongs to does not affect the principal component analysis calculation. By contrast, an sPLS-DA dimensionality reduction summarizes the differences between groups and calculates the parameters that contribute most to these differences. The "sparse" in sPLS-DA removes parameters that do not contribute to differences. For the sPLS-DA construction, M-fold validation was used. The Mahalanobis distance was calculated, and at least three components were generated for each set of comparisons during sPLS-DA construction. To calculate differences between groups, a PERMANOVA was done using the R package "vegan." 48 Permutation tests are powerful nonparametric tests, as they assume neither normality of distribution nor homogeneity of variance, and only assume data are exchangeable. 49 Permutation tests randomly resample data without replacement, which allows them to accommodate small sample sizes. 50 The same metaclusters selected by sPLS-DA were used for the PERMANOVA. Data were then scaled (to allow balanced comparisons between parameters), with the Euclidean distance being calculated between points. A total of 4999 permutations were performed to generate P-values (for a total of 5000 tests), to provide power and confidence for a = 0.01. 51 For pairwise comparisons the package "pairwiseAdonis" was used with Holm's correction for multiple comparisons. 52 For comparisons of B-cell subset/cluster levels (either proportions or cell counts) between all five groups [non-MS controls, patients with untreated MS (prior), and patients with MS post-1, post-2 alemtuzumab and relapse], a PERMANOVA was done followed by pairwise comparisons with Holm's correction as discussed above. Prior, post-2 and relapse groups were compared with non-MS controls (for three comparisons). When comparing between patients with MS before and after treatment, a linear mixed-effects model was calculated using the "lme4" package. 53 Individual patients were considered random effects for a repeated measures test that accommodated missing values, as not all patients had all available timepoints, while timepoints (prior, post-1, post-2) were fixed effects. A total of 4999 permutations were then run using the "permanova.lmer" function as part of the "predictmeans" R package to calculate P-values. [54][55][56] Five multiple comparisons were made: prior to post-1, post-2 and relapse; and post-1 to post-2 and post-2 to relapse. The functions "permmodels" and "predictmeans" (also part of the "predictmeans" package) generated 4999 permutations to calculate P-values with Holm's correction. A similar linear mixed-effects model was used when comparing between median signal intensities of minor B-cell subsets among nine non-MS controls. Multiple comparisons with Holm's correction were done when comparing between all minor subsets of the same major B-cell population, for a total of 34 combinations.
For comparisons between patients with active or inactive disease, and patients with and without prior treatments, a permutation t-test was done using the R package "RVAideMemoire", 57 using 4999 permutations.
All plots were generated using the R package "ggplot2". 58 Single-cell RNA sequencing data analysis Single-cell RNA sequencing data were obtained from Gene Expression Omnibus (accession number GSE133028). 11 Data included single-cell RNA sequencing data from PBMCs of healthy individuals and patients with RRMS (cohort GPL20301), and total RNA sequencing data from sorted B-cell subsets (cohort GPL27644). Single-cell RNA sequencing data were processed as described by Ramesh et al., 11 providing filtered single-cell gene count matrices. Data were then log normalized using "NormalizeData" as part of the "Seurat" package in R. 59 This cohort included PBMCs from healthy individuals (n = 3), and patients with neuromyelitis optica (n = 1), RRMS (n = 12), uveitis (n = 1) and clinically isolated syndrome (n = 2). All 175 895 cells underwent FlowSOM clustering and FIt-SNE dimensionality reduction using the following genes to differentiate B cells from other major cell subsets (adapted from Ramesh et al. 11 ): APOE, C1QA, C1QB, C1QC, CCR7, CD3E, CD3D, CD4, CD8A, CD8B, CD14, CD19, CD40, CD40LG, CD79A, CD79B, CST3, FCER1A, FCGR3A, GNLY, HEMGN, IL7R, LILRA4, LYZ, MS4A1, MS4A7, NKG7, PPBP, TRAC. FIt-SNE plots were run with a "perplexity" of 1750 and "learning_rate" of 15 000, based on Kobak and Berens. 60 B cells were identified and extracted for analyses. Only B cells from healthy controls and patients with RRMS were compared. Mean expression levels for each B cell were calculated and compared between healthy controls and patients with RRMS; 11/12 patients with RRMS were untreated, with 1/12 having previously received steroids. A permutation t-test was done using the R package "RVAideMemoire," 57  Total RNA sequencing data of B-cell subsets were processed as described by Ramesh et al., 11 providing reads per kilobase per million for each sample. PBMCs were taken from six patients with untreated RRMS. PBMCs were then flow cytometry sorted into five B-cell subsets: na€ ıve (IgD + CD27 À ), double negative (IgD À CD27 À ), unswitched memory (IgD + CD27 + ), switched memory (IgD À CD27 + ) and plasmablasts/plasma cells (IgD À CD27 hi ). Provided reads per kilobase per million were compared between groups using a permutation of linear mixedeffects model and multiple comparisons with Holm's correction (10 in total) as described above. Genes chosen for comparisons were the same as those above.