Transcriptional states of CAR-T infusion relate to neurotoxicity - lessons from high-resolution single-cell SOM expression portraying

Abstract

Anti-CD19 CAR-T cell immunotherapy is a hopeful treatment option for patients with B cell lymphomas, however it copes with partly severe adverse effects like neurotoxicity. Single-cell resolved molecular data sets in combination with clinical parametrization allow for comprehensive characterization of cellular subpopulations, their transcriptomic states, and their relation to the adverse effects. We here present a re-analysis of single-cell RNA sequencing data of 24 patients comprising more than 130,000 cells with focus on cellular states and their association to immune cell related neurotoxicity. For this, we developed a single-cell data portraying workflow to disentangle the transcriptional state space with single-cell resolution and its analysis in terms of modularly-composed cellular programs. We demonstrated capabilities of single-cell data portraying to disentangle transcriptional states using intuitive visualization, functional mining, molecular cell stratification, and variability analyses. Our analysis revealed that the T cell composition of the patient's infusion product as well as the spectrum of their transcriptional states of cells derived from patients with low ICANS grade do not markedly differ from those of cells from high ICANS patients, while the relative abundancies, particularly that of cycling cells, of LAG3-mediated exhaustion and of CAR positive cells, vary. Our study provides molecular details of the transcriptomic landscape with possible impact to overcome neurotoxicity.

Keywords: CAR-T cell immunotherapy; bioinformatics workflow; data portraying; single-cell transcriptomics; transcriptional states.

Figure 1

Overview of the downsampling and upscaling processes: (A) In short: Single-cell RNA sequencing data is reduced to meta-cell data. After self-organizing map (SOM) training using these meta-cells, data is upscaled to single-cell resolution. (B) In detail: scRNAseq data of 133,405 cells was used to generate tSNE and cell clusters using standard Seurat preprocessing workflow. The resulting clusters were then further divided in a refinement clustering step, resulting in 1,486 meta-cells. Expression data of these meta-cells were used for SOM training followed by definition of eleven expression modules A–K Finally, analysis is upscaled to single-cell level by calculating module expression data for each single cell, and stratification of the data by patterns of module activation (PATs).

Figure 2

Cellular identity decomposition using consensus markers: (A) Assignment of T cell subpopulations and t-SNE projection. Percentages refer to total number of cells in the data. (B) CAR positive cells decompose into CAR+ CD4 and CD8 cells, respectively. A small population of ICANS-associated cells (IAC) with myeloid characteristics (14) segregate in a separate cluster as indicated. (C) CAR expression as a function of CD8 expression in CD3+ T cells reveals that double positive cells express both markers, while single positive and negative populations distribute along the y- and x-axis, respectively. Proportions given in the figure relate to the total number of 132,236 CD3+ T cells in the data set.

Figure 3

Expression landscape of T cells: (A) Flow chart of single-cell data portraying workflow (see Supplementary Figure S5 for more details). (B) Expression portraits of most abundant subpopulations. Red and blue pixels represent meta-genes over- and under-expressed in the respective subpopulation, respectively, meta-genes colored in green show no differential expression. (C) Module definition and functional annotation. (D) Mean module differential expression averaged over all cells in the three main subpopulations, and grouped by subpopulation and by cell cycle phase, respectively. (E) Mapping of T cell and immune cell subpopulation makers.

Figure 4

Segregation of Treg, Th1 and Tc subpopulations according to cell cycle phase: (A) Relative amounts of cells in G1, G2M and S phase, respectively. (B) Module expression grouped by subpopulations and cell cycle phase. Blue dots represent under-expressed modules, red dots over-expressed ones. (C) Corresponding expression portraits. (D) Distribution of module B expression in all cells shows bimodal character according to resting (G1-phase) and cycling (S- & G2M-phase) cells, respectively. (E) Module localizations in the map and hierarchical clustering of the PATs. * denotes that the PATs are not yet split into major and minor PATs. (F) The overview heatmap shows combinations of activated modules (PATs; major and minor PATs are summarized in this plot). Rows represent the subpopulations and cell cycle phases, columns represent the PATs clustered according to similarity of their average module expression. The numbers in the map show –log10 p-values derived from Fisher’s exact test. CD8+ cells show activation of module J related to LAG3-mediated T cell exhaustion partly together with cell cycle module B.

Figure 5

Module activation patterns (PATs) of activated expression modules reveal modular combinatorics: (A) Frequency of the module combinations detected in the cells. Major PATs collect samples with all modules exceeding the 1σ expression threshold, minor PATs those with all modules >0.5σ, but some <1σ (see also example portraits). (B) The PAT map is generated using t-SNE on PATs’ average expression values. PATs enriched in the three major T cell subpopulations and cell cycle phases are highlighted (see panels (C–E). Red and blue areas include cycling (B-type) and LAG3-exhausted (J-type) PATs, respectively. (C–E) t-SNE map of PATs enriched in all CD4+ Treg, Th1 and CD8+ Tc, respectively. Size of the dots scales with enrichment (–log10 p-value in Fisher’s exact test, data is given as Supplementary File 2 ).

Figure 6

Overview of subpopulation expression portraits and relative abundances. CD8+ T cells were hierarchically stratified by CAR-status, by cell cycle activity as seen by module B expression, and by ICANS group (from top to bottom). Pairwise comparison of the ICANS portraits in the four main branches reveals virtually identical expression patterns between the low and high ICANS groups in each of the branches. Contrarily, the relative amounts of cells in the low and high ICANS groups show marked differences especially in cycling CAR-positive cells (right main branch) meaning that high ICANS associates with a roughly three times reduced fraction cycling CAR+ cells. This difference is reduced in non-cycling CAR+ and cycling CAR- cells and reverses in non-cycling CAR- cells.

Figure 7

T cell subpopulations and expression patterns stratified by neurotoxicity grade: (A) Total cell numbers and relative amount of CD4+ Treg, Th1, and CD8+ Tc cells observed in male patients with ICANS grade 0-2 and 3-4, respectively. P-values were computed using Wilcoxon rank-sum test. (B) Maps of enriched PATs in cells grouped by ICANS grade. T cell subpopulations are highlighted according to Figure 5 . (C) Fraction of cells in each male patient classified into the particular PATs (only PATs with p-value <0.1 in Wilcoxon rank-sum test are shown). Bar lengths represent the mean percentage for the two ICANS groups, the dots represent the individual patients. (D) Virtual PAT flow between ICANS groups. PATs on the green colored end of the flow are more frequent in ICANS grade 0-2 patient cells, those on the red colored end in ICANS grade 3-4. Width of the flow bars scale with the virtual flow.

Figure 8

Expression patterns and relative abundance of CAR-positive and –negative T cells: (A) Relative amount of CAR+ cells in the subpopulations, respectively. (B) Module expression values grouped by subpopulations and CAR status. Blue dots represent under-expressed modules, red dots over-expressed ones. (C) Corresponding expression portraits. (D) Maps of enriched PATs in CAR-positive and –negative subpopulations. (E) PAT flow between CAR-positive and –negative cells. (F) Fraction of CAR-positive cells in each male patient grouped by T cell subpopulation. Bar lengths represent the mean percentage for the two ICANS groups, the dots represent the individual patients. p-values were computed using Wilcoxon rank-sum test. (G) Relative amount of CAR-positive CD8+ T cells in male (left frame) and female patients (right frame) grouped by ICANS grade. p-values were derived from linear regression model.

Figure 9

Portraying of single-cell transcriptome landscapes illustrates that neurotoxicity associates with decreasing cycling activity, amount of CAR+ cells, and expression of modules B (cell cycle genes) and J (exhaustion related genes LAG3 and TIM3). The composition of the infusion product (IP) regarding different T cell types is virtually invariant. The figure shows mean expression portraits of all single T cells stratified by ICANS group, cell cycle phase (S & G2M versus G1), CAR status, and expression of modules B and J, respectively.