Single-cell analysis provides novel insights into elements of CAR T cell toxicities and response.
Despite the tremendous success of therapies with T cells engineered to express chimeric antigen receptors (CARs) targeting the signal-transduction receptor CD19 for relapsed and/or refractory large B cell lymphoma1,2,3, there remain opportunities for improving response rates, extending remission durability and reducing toxicity. Identifying elements of the infusion product and exploring the etiologies of CAR T cell toxicities that affect patient outcomes is an active area of investigation. Two studies, one in Nature Medicine4 and one in Cell5, recently used multi-dimensional analyses to investigate the cellular signature of CAR T cell products to obtain information of response and toxicity4, and also to identify previously unknown populations of cells that might contribute to these effects5 (Fig. 1).
CAR T cells are T cells genetically engineered to combine the extracellular domain of antibody single-chain variable fragment with the intracellular domain of T cells signaling; this therefore merges antibody-based specificity with T cell cytotoxicity to allow major histocompatibility complex–independent targeting of tumor antigens6. CAR T cells are typically generated from lymphocytes derived from autologously collected apheresis samples; thus, the composition of CAR T cells is uniquely personalized to the patient and incorporates vast heterogeneity, which may contribute to the wide presentation of cytokine-release syndrome, neurotoxicity (immune effector cell–associated neurotoxicity syndrome) and CAR T cell efficacy.
Initial analyses of patients’ apheresis samples and CAR T cell products, largely by flow cytometry, have identified key phenotypic markers associated with a patient’s response. Specifically, markers that demonstrate increased T cell exhaustion, which leads to a progressive decrease in T cell potency, have been associated with dysfunctional responses7. Conversely, the presence of naive or early memory T cells that may have better T cell potency correlates with an improved clinical response8, and depletion of these subtypes results in poor T cell expansion9. These studies highlight the implications of T cell phenotypes in association with anti-tumor efficacy. Single-cell analysis platforms, such as single-cell RNA sequencing (scRNAseq), facilitate deeper analysis, which can potentially lead to the identification of small, previously underappreciated populations that may be clinically relevant and provide insight into the molecular details of why some people respond or have more-severe toxicities, while others do not.
In their study reported in Nature Medicine, Deng et al. performed whole-transcriptome scRNAseq of infusion products (including CAR T cells and non-transduced T cells) for 24 patients with large B cell lymphoma. In their study, at 3 months after infusion, 14 patients had a partial response or progressive disease, 9 patients had a complete response, and 1 patient was not eligible for analysis4. They found that patients with a partial response or progressive disease showed enrichment for exhausted CD8+ T cells, and patients who achieved complete response showed enrichment for memory T cells (Fig 1). Moreover, in those with disease at a higher stage, there was reduced expression of markers consistent with memory T cells, which has also been associated with worse outcomes9,10,11.
A secondary analysis using a cell-free DNA assay to identify early molecular responses to therapy demonstrated that a decrease of more than fivefold in the detection of tumor by cell-free DNA on day 7 after infusion (relative to the amount on the day of infusion) was associated with an improved response at 3 months. By using the scRNAseq dataset, the authors identified that an exhausted T cell phenotype was more abundant in suboptimally responding patients who did not meet this threshold than in those who did, in further support of the correlation of exhausted T cells with a poor response.
Finally, they used their scRNAseq database to identify associations between the CAR product and toxicity. They identified a very small but significant cell population (254 cells) in the infusion product that was associated with more-severe neurotoxicity (Fig. 1). On the basis of gene-expression and single-sample gene set–enrichment analysis, these cells seemed to be myeloid in origin and expressed important cytokine and chemokine markers known to be associated with neurotoxicity related to cytokine-release syndrome, including IL-1β and CXCL8 (IL-8)12. Further study of the role of this cell population is warranted to assist efforts to mitigate neurotoxicity.
While the efforts by Deng et al. shed some light on neurotoxicity4, its etiology has remained largely elusive and has been postulated to be related to disruptions in the blood–brain barrier (BBB)13. Accordingly, in their study published in Cell, Parker et al. explored scRNAseq datasets from human brain to potentially identify on-target off-tumor causes of neurotoxicity in therapies using CD19-targeted CARs (CD19 CARs)5. They found that CD19 was expressed on mural cells, which line blood vessels and maintain the BBB. Subsequently, they demonstrated that mouse CD19 CARs disrupted the BBB by targeting CD19+ mural cells in mice, providing further insight into a biological basis for CD19 CAR–associated neurotoxicity (Fig. 1). They subsequently evaluated differentially expressed genes to distinguish degrees of expression between B cells versus brain pericytes and endothelial cells, which could provide insight into improving the B cell targeting and optimize safety.
As CAR T cells come to the therapeutic forefront, it is essential to understand the limitations and strengths of these treatments for patients, specifically by identifying features of CAR T cells or their targets that affect outcomes. As shown here, the concurrent explosion of high-dimensional technologies will enhance insight into the biology of CAR T cells. However, as multi-parametric analysis platforms become increasingly accessible, sample-handling, sample-processing and data-analysis pipelines must be established in the context of these novel therapies to facilitate such studies.
Deng et al. established a robust pipeline for collecting products from patients that facilitates real-time processing and enabled them to carry out several downstream analyses4. Additionally, they employed methods to enhance the identification of cell populations, an approach that should be considered and integrated into single-cell high-dimensional sample-processing and sample-analysis pipelines for future research. The clinical implications of their results are that real-time analysis, such as analysis of tumor-associated cell-free DNA at day 7, may provide early identification of those at risk of a poor response, for whom optimization strategies could be implemented. In parallel, the findings by Parker et al.5 highlight the power of incorporating a multimodal evaluation of high-dimensional analyses to validate findings, the need for preclinical modeling to understand the biology associated with these findings, and the power of combining large datasets to identify potential targets for immunotherapy optimization.
Importantly, not only are findings from both of these studies4,5 biologically aligned with previously established principles of the response to CAR T cell therapy and linked neurotoxicities, but also they contribute further to the understanding of this field, are hypothesis generating and are clinically relevant. The next step of combining such analyses with functional and phenotypic assessments at multiple relevant time points for patients will be invaluable for identifying the most desirable and functional cell populations in CAR products and target tissues, and may provide hints for modifications that could optimize the response to CARs. These studies not only will expand understanding of the biology of CAR T cells in patients but also will improve the ability to predict clinical response and toxicity and will inform the next generation of CAR T cell–based therapies.
Locke, F. L. et al. Lancet Oncol. 20, 31–42 (2019).
Schuster, S. J. et al. N. Engl. J. Med. 377, 2545–2554 (2017).
Neelapu, S. S. et al. N. Engl. J. Med. 377, 2531–2544 (2017).
Deng, Q. et al. Nat. Med. https://doi.org/10.1038/s41591-020-1061-7 (2020).
Parker, K. R. et al. Cell 183, 126–142.e117 (2020).
Eshhar, Z. in Therapeutic Antibodies. Handbook of Experimental Pharmacology Vol. 181 (eds. Chernajovsky, Y. & Nissim A.) 329–342 (Springer, Berlin, Heidelberg, 2008).
Finney, O. C. et al. J. Clin. Invest. 129, 2123–2132 (2019).
Fraietta, J. A. et al. Nat. Med. 24, 563–571 (2018).
Das, R. K., Vernau, L., Grupp, S. A. & Barrett, D. M. Cancer Discov. 9, 492–499 (2019).
Sommermeyer, D. et al. Leukemia 30, 492–500 (2016).
Locke, F. L. et al. Blood Adv. 4, 4898–4911 (2020).
Norelli, M. et al. Nat. Med. 24, 739–748 (2018).
Gust, J. et al. Cancer Discov. 7, 1404–1419 (2017).
The authors declare no competing interests.
About this article
Cite this article
Ramakrishna, S., Shah, N.N. Using single-cell analysis to predict CAR T cell outcomes. Nat Med 26, 1813–1814 (2020). https://doi.org/10.1038/s41591-020-01157-w