High-dimensional immune profiling identifies a biomarker to monitor dimethyl fumarate response in multiple sclerosis

Figures

• 

  Fig. 1.. DMF treatment changes the peripheral immune profile of MS. (A) Blood samples and clinical data from a cohort of MS patients (n = 31, as well as two validation cohorts) before the initiation of DMF therapy and at 3- and 12-mo follow-up were longitudinally collected and analyzed for immune profile by mass cytometry and SNPs by DNA microarray. (B) UMAP representation of analyzed peripheral blood immune cells with overlaid color code as categorized by FlowSOM-defined clusters. (C) Immune compartments as defined by FlowSOM-defined clusters. Mean expression profile of all analyzed parameters in each defined population. (D) Longitudinal frequencies of the immune populations in blood in relative numbers. (E) Longitudinal cell counts of the main immune populations in blood in absolute numbers.

• 

  Fig. 2.. DMF treatment primarily affects mature T and B cells. (A) UMAP representation of B cell and Th and Tc cell subsets with overlaid color code as categorized by FlowSOM-defined clusters. (B) The change of density (blue indicating decrease and red indicating increase during follow-up) within the UMAP representation was computed via a Gaussian kernel density estimation for change between T1 and T3. (C) Expression maps of representative markers of lymphocyte maturation in the UMAP representation. (D) Frequencies of established maturation cell types of Th cells, Tc cells, and B cells at time points T1 to T3 visualized in area plots. (E) Radar chart depicting the change of mean cytokine expression (T3/T1) in each cell type for CD4 T cells, CD8 T cells, and B cells. (F) Radar chart depicting the change of mean cytokine and chemokine receptor expression (T3/T1) in each cell type for CD4 T cells, CD8 T cells, and B cells.

• 

  Fig. 3.. Automated multipanel analysis with CellCnn identifies an inflammatory Tem phenotype associated with DMF therapy. CellCnn analysis determined the core immune features between T1 and T3 from our patient cohort, termed the “treatment-associated phenotype” (TAP). (A) The UMAP algorithm was used to depict the identified cell filter in the conventional (green) and stimulated (red) panels. (B) Frequency of selected cells within the Th compartment at T1, T2, and T3, and the relative fold changes compared with baseline. (C) The overlap between both cell filters is estimated using silhouette analysis. The black dashed line represents the silhouette score. (D) Frequency of selected cell types within the treatment-associated phenotype at T1 in the stimulated panel and in the conventional panel. (E) Expression patterns of the five key discriminant markers between the treatment-associated phenotype and the reference cell population for the stimulated panel. Distance between patterns for each marker quantified by the Kolmogorov–Smirnov (KS) test. (F) Expression patterns of the five key discriminant markers between the treatment-associated phenotype and the reference cell population for the…

• 

  Fig. 4.. CXCR3 effector-memory T cell population correlates with DMF treatment response. Longitudinal CellCnn analysis determined the characterizing immune features of the therapy response from our patients’ cohort, as defined by relapses and/or new/active MRI lesions, termed the “response-associated phenotype” (RAP). (A) The UMAP algorithm was used to depict the identified response-associated (RAP, purple) and DMF treatment–associated cell filters (TAP, pink). (B) Frequency of the response-associated cell filter in patients with and without signs of disease activity, arranged for all three time points. (C) The silhouette analysis suggests the overlap of response-associated and DMF treatment–associated cell populations. The black dashed line represents the silhouette score. (D) Frequency of selected cell types within the response-associated phenotype at T1 in the stimulated panel. (E) Expression patterns of key discriminant markers between the response-associated population (purple) and the reference cell population (gray). (F) ROC curve depicting the accuracy of predicting disease activity at T3 for absolute lymphocyte counts at T3, NfL at T3, and response-associated population frequency…

• 

  Fig. 5.. Two independent cohorts and serum neurofilament levels confirm the algorithm-identified DMF-associated immune phenotype. (A) One cohort of 13 patients was analyzed in an analogous setup to the mass cytometry cohort by spectral flow cytometry. Another cohort of 4 patients with samples from untreated, DMF-treated, and natalizumab (NAT)-treated time points was assessed by conventional flow cytometry. (B) Representative illustration of cytokine and chemokine receptor staining on Tem subpopulations. (C) ROC curve depicting the accuracy of predicting disease activity at T3 using Tem frequency at T3, RAP frequency (within Tem) at T3, and NfL levels at T3. Bars illustrate the predictive accuracy and SD at intragroup cross-validation. (D) Frequency of the response/treatment–associated phenotype (as identified by CXCR3+ TNF+ IFNγ+ GM-CSF+ Tem cells) in patients with and without signs of disease activity at time point T3 in the flow cytometry cohort. (E) Mean relative change between T1 and T3 of cytokine expression in Tem. Each radius arm represents one cytokine and purple color indicates individuals with disease activity. (F) Frequency of the response/treatment–associated phenotype (as…