Endothelial dysfunction and proinflammatory state determine severe hematotoxicity and inferior outcome of CAR-T therapy

Abstract

Hematotoxicity and infections are the main drivers of non-relapse mortality after chimeric antigen receptor (CAR)-T therapy. Consequently, reliable predictive biomarkers are highly needed to improve risk assessment and optimize patient management. In this study, we applied the immune-related adverse outcome pathway concept to delineate key events and risk factors of CAR-T-associated hematotoxicity. To identify predictive biomarkers, we performed flow cytometry and multiplex assays before and early after CAR-T infusion on 78 patients (ide-cel n = 31; axi-cel n = 24; and cilta-cel n = 23) undergoing CAR-T therapy. Severe hematotoxicity was linked to endothelial dysfunction, as evidenced by reduced levels of ANG1, soluble selectins, and increased soluble VCAM-1 (sVCAM-1) early after CAR-T infusion. Increased sVCAM-1, reflecting endothelial dysfunction, elevated soluble IL-2R (sIL-2R), indicating a proinflammatory state, and high tumor burden before lymphodepletion were key risk factors for CAR-T-associated hematotoxicity. Patients with elevated sVCAM-1 and sIL-2R at baseline (pre-lymphodepletion) exhibited significantly reduced overall survival (OS) (sVCAM-1; P = 0.0009), prolonged Grade 4 neutropenia (sVCAM-1; 12.1 vs. 6.0 days; P = 0.0016), more aplastic neutrophil recovery (5% vs. 30%; P = 0.007), and more severe infections (22.4% vs. 55%; P = 0.011). Baseline sIL-2R and sVCAM-1 demonstrated robust predictive value for prolonged neutropenia, severe infections, and mortality independently of key clinical variables such as the underlying disease and CAR-T product. Integration of these markers improves existing models and can help to refine risk assessment and guide individualized patient management in CAR-T therapy.

Figure 1

Incidence and severity of hematologic toxicities after chimeric antigen receptor (CAR)‐T. (A–D) Blood counts of neutrophils (A), thrombocytes (B), hemoglobin level (C), and lymphocytes (D) before start of lymphodepletion and after CAR‐T infusion shown as spaghetti plots depicting the individual course of every single patient (thin lines) as well as the best‐fit line from local polynomal regression (bold line) and its 95% confidence interval (dashed colored lines). The dashed gray lines show the border of respective Grade 3 cytopenias. (E) Mean cumulative duration of Grade ≥3 neutropenia, thrombocytopenia, anemia, and lymphopenia for ide‐cel, axi‐cel, and cilta‐cel patients. Whiskers indicate standard error of the mean (SEM). Statistical significance of the duration of cytopenias between the different CAR‐T products was tested by Kruskal–Wallis and Dunn's multiple comparison test. (F–H) Sankey diagrams showing the time course from early immune effector cell‐associated hematotoxicity (ICAHT) grade to late ICAHT grade for ide‐cel (F), axi‐cel (G), and cilta‐cel (H). NA = not available. Statistical significance of the frequency of early and late ICAHT between the different CAR‐T products was tested by Fisher's exact test. Of the three ide‐cel patients not available (NA) for late ICAHT classification, two had progressive disease, and one was lost to follow‐up on Day 82 with only one visit after Day 30. Of the five axi‐cel patients not available (NA) for late ICAHT classification, four had progressive disease, and one patient deceased due to severe infection (early ICAHT Grade 3). The one cilta‐cel patient not available (NA) for late ICAHT classification deceased of intracranial hemorrhage with only one visit after Day 30.

Figure 2

Post‐infusion factors associated with severe hematotoxicity after chimeric antigen receptor (CAR)‐T. (A) Spearman correlation analysis of the influence of cytokines and flow cytometry markers early after CAR‐T infusion (Days 3–10; BS3) and cumulative Grade ≥3 anemia, Grade ≥3 thrombocytopenia, Grade ≥3 lymphopenia, and Grade ≥4 neutropenia. The Spearman correlation coefficient (r = scalebar) and the respective P values are depicted in the heatmap with red indicating positive correlation and blue indicating negative correlation (*P < 0.05, **P < 0.01, and ***P < 0.001). (B–H) Univariate analysis of the influence of soluble E‐selectin (B), soluble P‐selectin (C), VEGF‐A (D), and ANG1 (E) on prolonged neutropenia (red), ANG1 (F), MMP‐1 (G), soluble P‐selectin (H), and soluble VCAM‐1 (I) on prolonged thrombocytopenia (blue) as well as ANG1 (J) and soluble VCAM‐1 (K) on prolonged anemia (green) after CAR‐T therapy, each depicted in comparison to cumulative duration. Best‐fit line (bold line) and 95% confidence intervals (dashed line) were calculated by simple linear regression. The Spearman correlation coefficient (r) and respective P‐ and q‐values (after adjustment for multiple testing) are depicted for each marker. Shown are selected markers that fulfilled a q‐value cutoff of <0.1. DC, dendritic cells; Mo, monocytes; pct, percentage.

Figure 3

Baseline factors associated with severe hematotoxicity after chimeric antigen receptor (CAR)‐T. (A) Spearman correlation analysis of the influence of cytokines and flow cytometry markers at baseline (before start of lymphodepletion; BS2) and cumulative Grade ≥3 anemia, Grade ≥3 thrombocytopenia, Grade ≥3 lymphopenia, and Grade ≥4 neutropenia. The Spearman correlation coefficient (r = scalebar) and the respective P values are depicted in the heatmap, with red indicating positive correlation and blue indicating negative correlation (*P < 0.05, **P < 0.01, and ***P < 0.001). Univariate analysis of the influence of baseline soluble VCAM‐1 (B), soluble IL‐2 receptor (C), baseline tumor burden (D), and soluble P‐selectin (E) on prolonged neutropenia (red). Univariate analysis of the influence of baseline soluble P‐selectin (F) and ANG1 (G) on prolonged thrombocytopenia (blue) and of soluble VCAM‐1 (H) on prolonged anemia (green). Univariate analysis of the influence of baseline monocytes (as percentage of HLA‐DR⁺ CD45⁺ leukocytes) (I) and CD8 T cells (as percentage of CD45⁺ leukocytes) (J) on prolonged lymphopenia (purple). The influence of baseline tumor burden was analyzed by simple logistic regression and quantified by the likelihood ratio test (G2). For the other variables, best‐fit line (bold line) and 95% confidence intervals (dashed line) were calculated by simple linear regression. The Spearman correlation coefficient (r) and respective P‐ and q‐values (after adjustment for multiple testing) are depicted for each tested marker. Shown are selected markers that fulfilled a q‐value cutoff of <0.1. DC, dendritic cells; Mo, monocytes; pct, percentage.

Figure 4

High baseline sVCAM‐1 and sIL‐2R before chimeric antigen receptor (CAR)‐T is associated with inferior outcome. Kaplan–Meier curves for overall survival (OS) (A, B) and progression‐free survival (PFS) (D, E) for patients with high (red) and low (green) sVCAM‐1 (A, D) and sIL‐2R (B, E) at baseline (before start of lymphodepletion; BS2). Patients with values ≥ 3 quartile of all patients were characterized as high (sIL‐2R ≥ 5398.8 pg/mL; sVCAM‐1 ≥ 2094.0 ng/mL). Dashed lines indicate the median of OS and PFS, respectively. Statistical significance was determined by log‐rank test. (C, F) Forest plots showing the results of multivariate Cox regression analysis on the influence of different covariates on OS (C) and PFS (F). HR/IPI high, high‐risk cytogenetics (multiple myeloma [MM]) or International Prognostic Index (IPI) ≥ 3 (diffuse large B‐cell lymphoma [DLBCL]); bulk/elev. LDH, presence of bulk or elevated lactate dehydrogenase (LDH) at baseline; sVCAM‐1 + sIL‐2R ≥ 3Q, sVCAM‐1 and sIL‐2R levels above the third quartile; AIC, Akaike information criterion.

Figure 5

High sIL‐2R and sVCAM‐1 before chimeric antigen receptor (CAR)‐T is associated with more prolonged neutropenia, a more aplastic neutrophil recovery and more severe infections. (A) Mean cumulative duration of Grade 4 neutropenia between patients with high and low sIL‐2R (10.3 vs. 6.6 days) or sVCAM‐1 (12.1 vs. 6.0 days) at baseline. Whiskers indicate standard error of the mean (SEM). Statistical significance was determined by Mann–Whitney U test. (B) Percentage of quick, intermittent, and aplastic neutrophil recovery phenotypes between patients with high and low sIL‐2R or sVCAM‐1 at baseline. (C) Distribution of infection grades (CTCAE grade) between patients with high and low sIL‐2R or sVCAM‐1 at baseline. (D) Overall response rates of patients with high and low sIL‐2R or sVCAM‐1 at baseline subdivided by percentage of patients achieving partial remission (PR)/very good partial remission (VGPR) and complete remission (CR). Statistical significance was determined by Fisher's exact test (B–D). (E) Peak CD4 and CD8 CAR‐T expansion (d0 until d14) per µL peripheral blood between patients with high and low sIL‐2R or sVCAM‐1 at baseline as determined by flow cytometry. Statistical significance was determined by Mann–Whitney U test. *P < 0.05, **P < 0.01, and ***P < 0.001. Patients with values ≥ 3 quartile of all patients were characterized as high (sIL‐2R ≥ 5398.8 pg/mL; sVCAM‐1 ≥ 2094.0 ng/mL).

Figure 6

Predictive value of baseline sIL‐2R and sVCAM‐1 on adverse outcomes after chimeric antigen receptor (CAR)‐T. Receiver operating characteristic (ROC) curves of the influence of baseline sIL‐2R (A) and sVCAM‐1 (B) on the occurrence of prolonged Grade 4 neutropenia of ≥7 days (31 events). ROC curves of the influence of baseline sIL‐2R (C) and sVCAM‐1 (D) on the occurrence of severe infections (≥Grade 3) (24 events). ROC curves of the influence of baseline sIL‐2R (E) and sVCAM‐1 (F) on the occurrence of death (21 events). For ROC curves, the respective area under the curve (AUC/AUROC) and P‐value are depicted. The concentration in brackets depicts the cutoff optimizing the Youden index. (G) Tabular summary of logistic regression analyses, including univariate models for baseline sIL‐2R, sVCAM‐1, CAR‐HEMATOTOX‐, EASIX‐, and mEASIX‐score, as well as multivariate models combining either sIL‐2R or sVCAM‐1 with CAR‐HEMATOTOX‐, EASIX‐, or mEASIX‐score. For each model, the estimate, P‐value, and Akaike information criterion (AIC) are reported. •P < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001.