Prediction of Organs Toxicity during CAR T-Cell Therapy By Using Cell-Free DNA Markers

Introduction

Chimeric antigen receptor T cell therapy is one of the rapidly developing and cutting-edge technologies for the treatment of cancer in recent years. However, CAR-T therapy can cause some side-effects due to cytokine storms and on-target off-tumor toxicity. The incidence of severe complications is about 30~40%, and serious ones can cause organ damage and even life-threatening. However, there are currently no biomarkers for overall assessment nor prediction of organ toxicity induced by CAR-T therapy. DNA methylation represents a fundamental epigenetic mark with tissue specific characteristics. Cell-free DNA with tissue specific methylation patterns will be released into peripheral blood when organ damage happens. If plasma cell-free DNA could be used for non-invasive and rapid prediction for the occurrence of fatal organ toxicity, then timely interventional treatment could be carried out (some CAR-T with switches can also be turned off in time), the survival rate of patients could be greatly improved.

Methods

We collected 157 peripheral blood samples from 42 patients receiving CAR-T therapy at day 1, day 4, day 7, day 10 after CAR-T cells infusion. Cell-free DNA were extracted from the plasma supernatant and underwent bisulfite conversion, and then was processed methylation single-strand library preparation and sequencing using DNBSEQ platform (MGI). Alignment was performed with the human reference genome (GRCh38/hg38) and PCR replicates were filtered with biobambam2 (v2-2.0.87)30. Tissue tracing analysis based on methylation density deconvolution method. Random forest model and LASSO regression were applied to screen for methylation sites that were diagnostic or predictive of organ toxicity. The logistic regression model was selected from the intersection of the two statistical methods to select the most suitable sequence for predicting organ toxicity.

Results

In this study, we identified some predictive biomarkers for ICANS, hepatotoxicity, and pulmonary toxicity separately. ICANS could be predicted by oligodendrocyte - derived cfDNA in plasma. When the concentration of oligodendrocyte - derived cfDNA was over 0.0035 ng/ml during day 1-8 after CAR-T infusion, the patient would be likely to suffer an ICANS in 4 days. The sensitivity and specificity were 85.7% and 71.4%, respectively. The area under the ROC curves (AUC) was 0.737. Besides, combination of hepatocyte-derived cfDNA and NK-derived cfDNA concentration during day 1-8 after CAR-T cells infusion could predict hepatotoxicity in advance of 1-19 days: predict probability (p) = exp(-2.655 + 0.262×NK + 0.151×Hepatocyte)/[1+exp(-2.655 + 0.262×NK + 0.151×Hepatocyte)]. Its sensitivity and specificity were 57.1% and 96.8%, respectively. The ROC-AUC of the predictive model was 0.763. In addition, concentration of alveolar-epithelium-derived cfDNA in plasma during day 1-8 after CAR-T cells infusion over0.180 ng/ml could predict pulmonary toxicity in advance of 1-7 days, with a sensitivity of 54.5% and a specificity of 87.5%, meanwhile its ROC-AUC was 0.712.

Conclusions

In conclusion, we realized organ tracing by plasma cell-free DNA methylation sequencing to predict organ toxicity caused by CAR-T cell therapy. This method is expected to be applied to the prediction of organ damage caused by diseases or other immunotherapies.