In-Depth Immunophenotyping With Mass Cytometry During TB Treatment Reveals New T-Cell Subsets Associated With Culture Conversion

Abstract

Tuberculosis (TB) is a difficult-to-treat infection because of multidrug regimen requirements based on drug susceptibility profiles and treatment observance issues. TB cure is defined by mycobacterial sterilization, technically complex to systematically assess. We hypothesized that microbiological outcome was associated with stage-specific immune changes in peripheral whole blood during TB treatment. The T-cell phenotypes of treated TB patients were prospectively characterized in a blinded fashion using mass cytometry after Mycobacterium tuberculosis (Mtb) antigen stimulation with QuantiFERON-TB Gold Plus, and then correlated to sputum culture status. At two months of treatment, cytotoxic and terminally differentiated CD8⁺ T-cells were under-represented and naïve CD4⁺ T-cells were over-represented in positive- versus negative-sputum culture patients, regardless of Mtb drug susceptibility. At treatment completion, a T-cell immune shift towards differentiated subpopulations was associated with TB cure. Overall, we identified specific T-cell profiles associated with slow sputum converters, which brings new insights in TB prognostic biomarker research designed for clinical application.

Keywords: T-cell; immunomonitoring; mass cytometry (CyTOF); tuberculosis; unsupervised analysis.

Figure 1

Experimental and analytical workflow. Peripheral whole blood samples were collected from active TB patients (n = 22) throughout treatment (T0: baseline. T1: T0 + 2 months. T2: end of treatment). After whole blood stimulation with Mtb antigens (TB2 and rmsHBHA) or with a negative control (NIL), total white blood cells were extracted. After palladium (Pd) barcoding for unique sample identification before multiplexing, T-cells were analyzed with a 29-marker mass cytometry panel. TB2, Qiagen QuantiFERON TB2 tube (ESAT-6 + CFP-10 + undisclosed CD8⁺ T-cell stimulating peptide pool); rmsHBHA, recombinant heparin-binding hemagglutinin obtained in Mycobacterium smegmatis; UMAP, Uniform Manifold Approximation and Projection; FlowSOM, self-organizing map.

Figure 2

Peripheral CD3⁺ T-cell unsupervised clustering and phenotyping. (A–E) FlowSOM automated clustering. The surface expression of lineage markers used for FlowSOM calculations was visualized in all CD3⁺ events (201,000 events from equally down-sampled files) regardless of timepoint or stimulation. FlowSOM enabled automated repartition of CD3⁺ events into 196 clusters according to the surface expression of selected lineage markers such as CD4 (A), CD8 (B), and CD45RA (C). Scales indicate arcsinh-transformed mass signal values. Clusters were automatically grouped into 18 meta-clusters of homogeneous phenotype, which were assembled into 12 canonical T-cell subpopulations in a supervised manner after meta-cluster phenotyping. This was performed with heatmap visualization of normalized, arcsinh-transformed median mass signal values for each surface marker (D). The proportions of the resulting T-cell subpopulations were visualized on the initial FlowSOM minimum spanning tree to control phenotyping consistency (E). (F, G) Reference mapping. Dimension reduction was performed with UMAP and overlayed with automatically determined FlowSOM clusters (F) and meta-clusters (G) to generate a phenotype reference map. Cluster labels were not displayed for legibility. CM, central memory; DN, double-negative CD4^-CD8^-; DN, double-positive CD4⁺ CD8⁺; EM, effector memory; MAIT, mucosal associated invariant T-cells; Tgd, gamma delta T-cells; Treg, T-regulators; TEMRA, terminally differentiated effectors re-expressing CD45RA.

Figure 3

Significant abundance changes in non-canonical T-cell subsets throughout TB treatment. FlowSOM cluster abundance was analyzed over time in unstimulated or Mtb-stimulated samples (TB2 or rmsHBHA). Only clusters within which significant abundance changes were detected were displayed. Number of matched data points per timepoint for all panels: NIL: n = 16. TB2: n = 18. rmsHBHA: n = 14. Data are represented as medians + interquartile range. (A–D). Significantly increased clusters at treatment completion (T2) compared to treatment initiation (T0). Clusters within which a significant increase was detected between T0 and T2 were visualized on the reference UMAP shown in Figure 3 (A). Cluster abundance quantification was was performed in unstimulated (B), TB2-stimulated (C) or rmsHBHA-stimulated samples (D). (E–H) Significantly decreased clusters at treatment completion (T2) compared to treatment initiation (T0). Mapping (E) and abundance quantification of clusters which increased between T0 and T2 in unstimulated (F), TB2-stimulated (G) or rmsHBHA-stimulated samples (H). DN, double negative CD4^- CD8^-; Tgd, gamma delta T-cells. Statistical analysis: Friedman rank sum test and Wilcoxon-Nemenyi-Thompson post-hoc for pairwise comparisons between non-independent observations at T0, T1, and T2. Exact, unadjusted p-values are indicated on the figures. Benjamini-Hochberg corrections for multiple comparisons were performed as an indication and were not used for cluster selection for phenotype analyses in order to minimize type II errors. Adjusted p-values did not reach significance. All adjusted p-values and complete test statistics are available in Supplementary Table 5 .

Figure 4

In-depth phenotyping shows differential involvement of effector and memory T-cells in cluster abundance changes during TB treatment. Mean marker expression levels were visualized using heatmapping for cell cluster which increased (orange color code) or decreased (green color code) throughout treatment. Each line represents one cell cluster. Scales indicate normalized mass signal intensity. Hierarchical clustering was performed based on marker expression levels, regrouping cell clusters of similar immunophenotypes. Black rectangles annotated from (A–D) indicate cell cluster subgroups with both similar abundance changes and similar immunophenotypes (i.e. at least same main T-cell subset).

Figure 5

Individual immunoprofiling confirms differential abundance of correlated subsets in cured patients after treatment. Cluster were stratified by type of significant abundance change: enrichment (A-C) or depletion (D-F) after treatment completion. (A, D) Pearson’s correlations were calculated on cluster abundance at T0 and displayed on a heatmap with hierarchical clustering. Clusters with similar immunophenotypes ( Figures 3 , 4 ) and positive correlation coefficients were grouped. Estimates of effect sizes are in Supplementary Tables 6, 7 . (B, C, E, F) The abundance of each subgroup was visualized. Each dot represents data for one patient. Statistical analysis: Friedman rank sum test. Subgroup A: data from rmsHBHA samples (n =14), clusters 74, 102, 160; p = 0.020, Fchisq = 5.4. Subgroup B: data from unstimulated samples (n =16), clusters 137, 154, 65, 82; p = 0.0013, Fchisq =10.3. Subgroup C: data from unstimulated samples, clusters 38, 54, 69, 28; p = = 0.0013, Fchisq = 10.2. Subgroup D: data from rmsHBHA samples, clusters 37, 70, 98; p = 0.032, Fchisq = 4.57. (F) For each subgroup, normalized mean marker expression levels were compared with similar T-cell subsets. (G–K) Manual gating analysis was performed to verify unsupervised results (representative plots, 500 to 1,000 events). Numbers indicate the percentage of gated cells among total CD3⁺ cells. Subgroup A: CD3⁺CD8⁺CCR7^-CD45RA^-CD7⁺Perforin⁺. Subgroup B: CD3⁺ CD4⁺CCR7⁺CD45RA^-CCR6⁺IL7Ra⁺CD27⁺CD40L⁺CD38⁺HLA-DR⁺. Subgroup C: CD4⁺CCR7^lowCD45RA^-CCR4⁺CCR6⁺ CD26⁺IL7Ra⁺CD7⁺CD27⁺. Subgroup D: CD4⁺CCR7⁺CD45RA^-CCR4⁺CCR6⁺CD26⁺IL7Ra⁺CD7+CD27⁺ CD38⁺.

Figure 6

Patients with slow microbiological culture conversion show decreased cytotoxic CD8⁺ and γδ enriched CD4⁺ naïve T-cell subsets before treatment initiation and after two months of treatment compared to fast converters. Fast converters (n = 18) were defined as patients with permanently negative M. tuberculosis culture after two months of treatment (T1), whereas slow converters (n = 4) were defined as patients with persistently positive cultures at T1. The abundance of all FlowSOM clusters at baseline was compared between fast and slow converters. CD4⁺ clusters were represented in red, CD8⁺ clusters in blue, and γδ T-cell clusters in green. Clusters which were significantly decreased (A, C) or enriched (B, D) at T1 in slow converters compared to fast converters were compared to the reference UMAP (E). Normalized, arcsinh-transformed mean marker expression levels were visualized (F). Each row represents one cluster. Scales indicate normalized mass signal intensity. Boxplot data represent medians + interquartile range. Statistical analysis: Only clusters within which differences passing a threshold of p<0.035 (Mann-Whitney U test) were represented. Exact, unadjusted p-values are indicated on the figures. Benjamini-Hochberg corrections for multiple comparisons were performed as an indication and were not used for cluster selection for phenotype analyses. Adjusted p-values did not reach significance. All adjusted p-values and complete test statistics are available in Supplementary Table 8 .

Figure 7

Non-lineage markers discriminate slow and fast responders within differentially abundant subsets. Principal Component Analysis (PCA) was performed on marker expression data from the clusters identified in Figure 6 , within 96 Mtb-stimulated samples matched at T0, T1, and T2 (TB2: 54 samples; rmsHBHA: 42 samples; see Supplementary Table 1 for sample number details). (A) Explanation of the variance between fast converters (25 samples at each timepoint) and slow converters (7 samples at each timepoint). Axes represent the principal components 1 (Dimension 1, Dim1) and 2 (Dim2). Percentages indicate their contribution to the total observed variance. Axis values represent individual PCA scores. Concentration ellipses correspond to 90% data coverage. (B) Contribution of cellular markers to the variance described by Dim1 and Dim2. Axis values represent marker PCA scores. Color codes represent broad marker functions. (C, D) Quantification of (B) for Dim1 (C) and Dim2 (D). Contributions of each marker are expressed as a percentage of the dimensions. The dashed line corresponds to the expected reference value if each marker contributed uniformly to the variance. Markers indicated in gray are below this reference value. (E, F) Distribution of individual PCA score values according to the culture conversion group at each timepoint, for Dim1 (E) and Dim2 (F). Wilcoxon Rank Sum Test. *p < 0.05. **p < 0.001. Exact p-values and test statistics are in Supplementary Table 9 .