Mass Cytometry of CSF Identifies an MS-Associated B-cell Population

Comparison of Blood and CSF Cells

We obtained CyTOF data from CSF of 25 control donors and 14 donors with MS, of which 2 had received rituximab and were analyzed separately. The resulting data were clustered with a benchmarked automated clustering routine; the clusters are displayed on the tSNE plot in figure e-1, links.lww.com/NXI/A386. First, we clustered all events together, without separating according to the anatomic compartment (blood vs CSF) or diagnosis. The clearest divisions corresponded to the 4 categories of T cells, B cells, NK cells, and myeloid cells (figure e-1, A–D), but the clustering approach enabled visualization of finer subgroups, such as the distinction within the T-cell population of CD45RA-expressing populations (clusters 13, 14, 15, and 16, figure e-1, A and C) vs CD45RO-expressing populations (clusters 1, 3, 4, 9, and 16, figure e-1, A and C). There were large differences in the abundances of these clusters between the blood and the CSF (figure e-1). In the T-cell compartment, cells in cluster 1 (expressing CD3, CD4, CD25, CD27, CD45RO, and CXCR3, i.e., memory T cells) were relatively more abundant in the CSF than in the blood. Conversely, clusters 13 and 14, also expressing CD3 and CD4 but characterized by positive CD45RA, negative CD45RO, and negative CD27, i.e., naive T cells, were more abundant in the blood. In the non–T-cell compartments, the most striking differences between CSF and blood were seen in clusters 5 and 7, of CD14-expressing monocytes, which were more abundant in blood (figure e-1E). This pattern of more naive T cells and innate cells such as monocytes in the blood than CSF matches previous reports.³ When analyzed this way, there were no significant differences between MS and control in any of the clusters in the CSF or blood, in line with previous findings.^3,18 However, because the blood cells in the pooled analysis were hundreds of times more numerous than the CSF cells, the definitions of the clusters were dominated by the blood cells, and therefore unlikely to be appropriate for categorizing the CSF cells, whose phenotypes are very different (figure e-1E).

CSF From Donors With MS Contains Characteristic B Cells

To extract cell phenotypes more representative of the CSF, we repeated the clustering using only the CSF cells and plotted it in the same way (figures 1, A–C). This revealed 2 populations with significantly different relative abundance in CSF from patients with MS vs controls. First, a naive CD8⁺ T-cell population (cluster 4 in figures 1 and 2), expressing CXCR5, CD25, CCR7, TIGIT, and CD45RA, was reduced in the CSF of patients with MS, albeit only significantly before correction for multiple comparisons (p = 0.039, FDR = 0.31). Second, of the 3 B-cell clusters (clusters 8, 9, and 13), the subpopulation characterized by CD49d, CD45RA, CD20, high CD27, CD69, and CXCR3 expression (cluster 8 in figures 1 and 2) was significantly increased in MS (p = 0.0012, FDR = 0.0192). Two of the 3 B-cell populations, including the MS-associated cluster 8, expressed CD20, and these were reduced in rituximab-treated patients, whereas the CD20-negative cluster 9 was not; however, the small number of rituximab-treated patients precluded statistical inference.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 Cell Types in CSF of Patients With MS or Other Diagnoses

(A) tSNE plot of cells from CSF. Data were clustered using k-means, and the clusters colored as shown in the key on the right of the figure, in a fashion exactly analogous to figure e-1, links.lww.com/NXI/A386. The characteristics of these clusters are shown in heatmap (C). The tSNE plot on the left shows data from cells from CSF of patients with non-MS diagnoses, the figure in the middle shows data from patients with MS, and the figure on the right shows data from patients with MS treated with rituximab. The red arrow in the bottom right corner of the middle tSNE plot indicates a cluster that is more abundant in CSF from patients with MS than in controls (see below). For this analysis, cells from CSF were clustered and analyzed independently, unlike in figure 1, enabling the detection of clusters that are specific to CSF, and too small to see against the background of the much larger dataset from blood. (B) Areas of the tSNE plots in (A) have been matched to the broad cell types myeloid, NK cells, T cells, and B cells. (C) Heatmap showing, for each of the clusters generated by k-means, the average intensity of each marker across all cells in the cluster. The colored bars at the top of the heatmap match the colors used in the tSNE plots in (A), and clusters corresponding to well-established cell types are shown on the right. Higher average expression of each marker is indicated with a red-brown color, and lower expression in blue.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2 Relative Abundances of Cell Types in CSF of Patients With MS or Other Diagnoses

For each donor, the relative abundance of the cluster (% of total cells) was plotted for each cluster described in figure 1. Cluster identity numbers are at the top of each boxplot, and the colors of the bars at the top of each boxplot match the colors used to mark the clusters in figure 1. Within each scatter column, each filled circle represents 1 donor. The diameter of the circles reflects the absolute size of the cluster in that donor. The 35 donors are separated into 3 groups: donors with other diagnoses than MS (control, light gray boxes on the left of each boxplot); donors diagnosed with MS, but not treated with CD20-depleting antibodies (dark gray box in middle); and donors diagnosed with MS and treated with rituximab (gray box on right). Data from 2 rituximab-treated patients are shown for interest without statistical inference. The asterisks mark the cluster in which there was a statistically significant difference in abundance between control and MS (generalized linear mixed model; **p = 0.0012, FDR = 0.0192). FDR = false discovery rate.

If this MS-specific B-cell population is pathogenic, preventing its access to the CNS with natalizumab ought to increase its frequency in peripheral blood. To test this, we longitudinally analyzed blood from patients before and after the start of natalizumab. The CXCR3+ HLADR+ CD45RA+ B-cell population increased within the lymphocyte population by 63% (figure e-2, links.lww.com/NXI/A387).

Cytokines, Chemokines, and B Cell–Related Molecules are Altered in MS CSF

To relate observed cellular characteristics to soluble signaling factors, we measured 296 proteins in CSF. First, we compared the concentrations of all analytes between donors with MS and controls. At an FDR cutoff of 5%, 162 analytes were differentially abundant in CSF between donors with MS or other diagnoses. The fold change levels of these differentially abundant solutes are shown as a volcano plot in figure 3 and as a heatmap in figure e-3, links.lww.com/NXI/A388. Several molecules were differentially abundant between MS and control whose binding partners were also different. These pairs of interacting proteins are compiled in table e-3, links.lww.com/NXI/A392. Most strongly increased in MS was the costimulatory receptor CD27, whose only known binding partner, CD70, was also more abundant. Other molecules differentially abundant at FDR < 10⁻⁵ can be classified into 4 categories: chemokines, cytokines, B cell–associated proteins, and neuronal signaling molecules. CCL4, CCL17, CCL28, and CXCL12 were positively associated with MS. Other chemokines including CXCL8, CXCL10, and CXCL13 also showed large fold changes in MS compared with controls, but the p values were also larger, reflecting greater heterogeneity within the MS group. Cytokines associated with MS were broadly of the Th2 type, including IL-5 (and its receptor), IL-33 (and receptor), as well as IL-13 and IL-21. We also observed higher levels of IL-4, IL-12 (and receptor), IL-18, IL-20, TNF-alpha, and TNF-beta. Three of the molecules most clearly different between MS and controls were noticeable because they are centrally involved in B-cell biology—TNFSF13B, ICOSLG, and FCRL2. TNFSF13B (tumor necrosis factor superfamily 13B, also called BAFF) is a critical survival factor for B cells,¹⁹ and was significantly less abundant in the CSF of donors with MS. Inducible T-cell costimulator (ICOS) is expressed by T cells (especially Th2 cells), and interaction with its ligand ICOSLG on B cells is costimulatory for both, leading to shedding of ICOSLG.²⁰ ICOSLG was less abundant in the CSF of donors with MS. The Fc-receptor-like family FCRL1-5 are principally expressed on B cells, especially FCRL2,²¹ which was more abundant in CSF from donors with MS. Finally, neuronal signaling molecules exhibited differences in abundance between MS and control CSF rivaling those of the immune-related molecules. For example, THY1 and NBL1 were significantly increased in MS, whereas a large number of neural signaling molecules including ROBO2 and EPHB6, were reduced in MS.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3 Volcano Plot of Soluble Markers in CSF From Donors With MS or Non-MS Diagnoses

The vertical axis is the log₁₀ of the p value for the comparison. The horizontal axis is the log₂ of the fold change. Names of analytes with FDR of less than 0.05 are printed beside the point. Red points correspond to analytes that are significantly more abundant in CSF from patients with MS, and blue points to analytes that are less abundant in MS. Analytes discussed in the text are indicated with arrows, whose directions and colors are explained on the right of the figure. FDR = false discovery rate.

To visualize the relationships between the soluble molecules and the clinical parameters, we subjected the Olink data to principal component analysis (PCA), whose major contributing factors are shown in figure 4A and plotted the individual donors on plots of the principal components PC1 and PC2 (figure 4B). Two findings emerge. First, the MS and non-MS CSF samples are displaced along a vector dominated by molecules involved in neuronal development and plasticity, such as ROBO2, EPHB6, CNTN5, UNC5C, RGMB, and MDGA1. Second, the molecules that we found to be strongly associated with MS, including soluble CD27, CXCL10, FCRL2, CHIT1, CD8A, IL-12, and TNFSF14 form a second axis, orthogonal to the neuronal plasticity vector. On this MS disease activity vector, the early stages of MS are associated with a shift toward the lower right of the plot (clinically isolated syndrome and relapsing-remitting forms), whereas samples from donors at later stages of disease are found closer to the non-MS controls (figure 4B). Samples from treated patients also appeared closer to the center of the plot, although comparison of treated vs untreated patients with MS did not reveal significantly different molecules.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4 Principal Component Analysis of Soluble Molecules Measured by PEA

(A) The contribution of each of the soluble molecules to the first 2 principal components PC1 and PC2 is shown by plotting each as a vector. For example, CXCL13 makes a large contribution to PC1, whereas HMOX1 mostly contributes to PC2. For clarity, only the 60 most strongly discriminating molecules are labeled. The broken green arrow pointing down and left indicates a hypothetical axis of neural plasticity, inferred from the preponderance of molecules involved in neural development, axonal signaling, and so on. The red arrow pointing down and right indicates a hypothetical axis of disease activity, inferred from the preponderance of inflammatory molecules, and the presence in this quadrant of early disease stages (see plot B). Candidate MS-associated, plasticity-suppressive chemokines CXCL1 and CXCL8 (see Discussion) are marked with green asterisks. (B) PCA sample scores plot, where each point represents one donor, and the color and shape of each point indicate the diagnosis, disease stage, and treatment of the donor, as shown in the legend below the plot. Green and red arrows are identical in meaning to (A). PEA = proximal extension assay.

Soluble CD27 and FCRL2 Correlate Strongly With MS-Associated B-Cell Population

To elucidate the relationship between these soluble molecules and the cellular phenotypes characterized by CyTOF, we examined the correlation between each of the differentially expressed solutes and the abundances of the differentially abundant cell clusters (figure e-4, links.lww.com/NXI/A389). This revealed a group of genes including CD27, FCRL2, CXCL10, CXCL13 TNFSF14, and lymphotoxin-alpha, almost perfectly overlapping with the genes of the MS disease axis shown in figure 4, whose expression was strongly positively correlated with the abundance of the MS-associated B cell population (cluster 8 in figures 1 and 2). Independently from this group, BAFF (TNFSF13B) was negatively correlated with cluster 8. Also using this hierarchical clustering of the soluble proteins, based on correlation with the cell-type abundances, a clear group emerges that resembles the neural plasticity axis from figure 4. Proteins of this group do not show a clear relationship with the abundance of the MS-associated B-cell population, reinforcing the inference that these 2 axes (disease activity and neural plasticity) are independent.