Fig 1: Increased cellular expression of chemokine receptors CCR2, CCR5, and CXCR3 in O-PML CSF(A) Feature plots depict expression level and distribution of CCL2, CCL3, CCL4, CCL5, CCL20, CXCL10, CXCL13, and CXCL16 in CSF of combined scRNA-seq datasets.(B) Quantifications of the proportion of cells positive for CCL2, CCL3, CCL4, CCL5, CCL20, CXCL10, CXCL13, and CXCL16 as percentage of mononuclear cells (MNCs) per sample in the CSF of IIH (n = 9), RRMS (n = 12), RRMS-Nat (n = 5), O-PML (n = 3), and VE (n = 5) patients are shown.(C) Feature plots depict expression level and distribution of CCR2, CCR5, CCR20, CXCR3, CXCR5, and CXCR6 in CSF of combined scRNA-seq datasets.(D) Quantifications of the proportion of cells positive for CCR2, CCR5, CCR20, CXCR3, CXCR5, and CXCR6 as percentage of cell in the lineage cluster per lineage and sample in the CSF of IIH patients, RRMS patients, RRMS-Nat patients, O-PML patients, and VE patients are shown. Boxplots show the IQR, with whiskers indicating IQR x 1.5, line denoting the median, and “+” indicating the mean. Statistical differences between IIH and O-PML were computed using Wilcoxon tests. In case of significant differences, statistical difference between IIH and RRMS and IIH and VE was also computed. ∗p < 0.05, ∗∗p < 0.01. IIH, idiopathic intracranial hypertension; RRMS, treatment-naïve RRMS patients; RRMS-Nat, natalizumab-treated RRMS patients; O-PML, non-MS patients at time of non-natalizumab-associated PML diagnosis; VE, viral encephalitis patients.
Fig 2: CCL3, CCL5, CXCL10, and CXCL13 are general inflammatory chemokines in CSF, while CCL2 is more specific for O-PML(A) Chemokine levels in CSF supernatant of RRMS (n = 8), RRMS-Nat (n = 6), Nat-PML (n = 6), and O-PML (n = 9) patients relative to CTRL (n = 8), computed by linear regressions of log2-transformed mean fluorescence intensity (MFI) from multiplex-bead assay. Colors indicate beta coefficients.(B) Chemokine levels in CSF supernatant of Nat-PML and O-PML patients relative to RRMS, computed by linear regressions of log2-transformed MFI from multiplex-bead assay. Colors represent beta coefficients.(C) Chemokine levels in CSF supernatant of Nat-PML patients relative to RRMS-Nat, computed by linear regressions of log2-transformed MFI from multiplex-bead assay. Colors indicate beta coefficients.(D) Venn diagram showing overlap of significantly increased chemokines in CSF of RRMS, RRMS-Nat, Nat-PML, and O-PML compared to CTRL. Linear regression models are described in detail in the methods. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. CTRL, patients with somatic disorder; RRMS, treatment-naive RRMS patients; RRMS-Nat, natalizumab-treated RRMS patients; Nat-PML, RRMS patients at time of natalizumab-associated PML diagnosis; O-PML, non-MS patients at time of non-natalizumab-associated PML diagnosis.
Fig 3: Dysregulation of chemokine networks in FRCs is linked to a diffuse growth pattern in lymphoma.a, Left, UMAP of LN stromal cells by entity. Right, percentages of rFRCs and rBECs (rLN n = 5, FL n = 6, DLBCL n = 8 patients). b, Differentially expressed genes between rLN-derived (n = 2,363 cells) and DLBCL-derived (n = 2,983 cells) FRCs (adjusted P < 0.05, log(fold change) > 0.5). c, Homeostatic (top) and inflammatory (bottom) chemokine expression in bulk data22. d, Pearson correlation of CXCL13 expression and CIBERSORTx-derived FDC fractions. e, CXCL13 plasma protein levels in FL (n = 18 patients) and DLBCL (n = 22 patients). Vertical lines indicate the mean per entity. f, Exemplary rLN and DLBCL mIF images, representative of n = 4 patients per entity. Scale bar, 50 μm. Dashed circles: CD21+ regions. g, mIF-derived CXCL13 and CXCR5 signals averaged across four adjacent pairs of CD21+ follicular and CD21− extrafollicular regions per sample. h,i, Spatial transcriptomics plots of FL-LN (h) and DLBCL-LN (i) cores colored by cell type and CXCL13–CXCR5 ligand–receptor (L–R) score. j, mIF-derived enrichment of CXCL13+ cells per cell type in DLBCL versus rLN/FL samples. Asterisks indicate P < 0.01; exact P values are provided in the source data. k, CXCL13 expression in CD8+ T cells (n = 21,268 cells)12. l, Percentage of CXCR5+ cells within CD3− fractions measured by flow cytometry (rLN n = 7, FL n = 24, DLBCL n = 18 patients). m, Migrated rLN- and DLBCL-derived B cells in the Transwell assay (mean ± s.d., n = 3 patients per condition). For c and d: tonsil n = 10, FL 1/2/3A n = 145, FL 3B n = 48, DLBCL n = 430 patients. For f, g and j: rLN n = 4, FL n = 5, DLBCL n = 4 patients. P values in a, c, e, g and l: two-sided Wilcoxon rank-sum test. P value in m: two-sided unpaired Welch’s t test. P values in j: two-sided Fisher’s exact test. P values in c, g and j were adjusted using the Benjamini–Hochberg method. Box plots: center line, median; box, interquartile range; whiskers, 1.5× the interquartile range; points, data values. FC, fold change; Ttox EM, effector memory cytotoxic T cells; hr, human recombinant.Source data
Fig 4: Spatial chemokine profiling identifies CD8+ effector memory T cells as an extrafollicular source of CXCL13 in DLBCL.(A) Representative triangle-thresholded mIF images showing CD21 (top panels) as well as CXCL13 (bottom panels) signal in rLN (n = 4), FL (n = 5) and DLBCL (n = 4) patient samples. Scale bar indicates 50μm. (B) UMAP representation of spatial transcriptomics data colored by cell type (color code depicted in panel C). (C) Heatmap showing scaled expression of key marker gene expression as well chemokines. (D) Stacked bar plot of FDC/FRC/rFRC fractions (top panel) and violin plot of CXCL13 expression in these subsets (bottom panel). (E) Spatial transcriptomics plots of FL (left panels) and DLBCL (right panels) tissue cores colored by cell type (left panels) alongside magnified views of chemokine expression. (F) Heatmap showing scaled expression of key T cell markers across CD8+ T cell populations as measured using mIF. (G) Box plot showing per-sample percentage of CXCL13+ cells (as determined using Otsu thresholding) among PD1+ CD8+ effector memory T cells in the mIF dataset. Boxes indicate the median (center line), interquartile range (bounds) and 1.5x interquartile range (whiskers). Data points represent patient samples. P-values were calculated using a two-sided unpaired Welch’s t-test. (H) Spearman correlation coefficients of key T cell markers across all CD8+ T cell populations12 (n = 21,268 cells). (I) Volcano plot showing differentially expressed genes comparing PD1+ CD8+ effector memory T cell populations12 (TTOX EM-II n = 2,018; TTOX EM-III n = 9,208 cells). Labels indicate cluster-identifying genes as well as CXCL13. Abbreviations: TTOX EM = effector memory cytotoxic T cells. For panels A, F, G: rLN n = 4; FL n = 5; DLBCL n = 4 patients. For panels B–E: FL n = 1; DLBCL n = 1 patient. Source data
Fig 5: Changes in chemokine expression profiles are associated with worse overall survival in DLBCL.a, UMAP representation of a microarray dataset22 with homeostatic (CXCL12, CXCL13, CCL19, CCL21) and inflammatory (CXCL9, CXCL10, CXCL11) chemokine expression values used as features for dimensionality reduction. Left, UMAP displaying pie charts (within each dot) that represent the k = 20 nearest neighbors, colored according to disease entity. Right, the same UMAP colored according to the mean expression of homeostatic (top) and inflammatory (bottom) chemokines. b, Bulk RNA-seq dataset29 stratified according to homeostatic chemokine expression into high (n = 519 patients) and low (n = 99 patients) groups using maximally selected rank statistics, shown as a dot plot (top) and a Kaplan–Meier curve of overall survival (bottom). The P value was calculated using the log-rank test. c, Forest plot summarizing log10-transformed hazard ratios (center), 95% confidence intervals (error bars) and Wald-derived P values estimated from univariate Cox proportional hazards models assessing the association between homeostatic chemokine expression and overall survival across five individual DLBCL bulk datasets22,28–31. d, Same UMAP as in a, colored by CIBERSORTx-derived FDC fractions. e, Bulk RNA-seq dataset29 stratified by FDC abundance into high (n = 255 patients) and low (n = 364 patients) groups using maximally selected rank statistics based on CIBERSORTx fractions, shown as a Kaplan–Meier curve of overall survival (left; log-rank test) and a scatter plot of log-transformed FDC fractions and CXCL13 expression (right; Pearson correlation). f, Forest plot summarizing log10-transformed hazard ratios (center), 95% confidence intervals (error bars) and Wald-derived P values estimated from Cox proportional hazards models assessing the association between FDC fraction and overall survival across five individual DLBCL bulk datasets22,28–31. For panels a and d: tonsil n = 10, FL 1/2/3A n = 145, FL 3B n = 48 and DLBCL n = 430 patients. OS, overall survival; HR, hazard ratio.Source data
Supplier Page from R&D Systems, a Bio-Techne Brand for Recombinant Human CXCL13/BLC/BCA-1 Protein