Dissecting the tumor-immune landscape in chimeric antigen receptor T-cell therapy: key challenges and opportunities for a systems immunology approach

Abstract

The adoptive transfer of genetically engineered Chimeric Antigen Receptor (CAR) T-cells has opened a new frontier in cancer therapy. Unlike the paradigm of targeted therapies, the efficacy of CAR T-cell therapy depends not only on the choice of target, but also on a complex interplay of tumor, immune, and stromal cell communication. This presents both challenges and opportunities from a discovery standpoint. Whereas cancer consortia have traditionally focused on the genomic, transcriptomic, epigenomic, and proteomic landscape of cancer cells, there is an increasing need to expand studies to analyze the interactions between tumor, immune, and stromal cell populations in their relevant anatomical and functional compartments. Here, we focus on the promising application of systems biology to address key challenges in CAR T-cell therapy, from understanding the mechanisms of therapeutic resistance in hematologic and solid tumors to addressing important clinical challenges in biomarker discovery and therapeutic toxicity. We propose a systems biology view of key clinical objectives in CAR T-cell therapy, and suggest a path forward for a biomedical discovery process that leverages modern technological approaches in systems biology.

Introduction

CAR T-cell therapy has seen exceptional success in several hematologic malignancies (1–5), marked by the first round of FDA approvals and Phase II trials (4,6). Yet these studies have also highlighted key clinical challenges including therapeutic resistance in a subset of patients, challenges in translation to solid tumors, and therapeutic toxicity (7–9). As clinical knowledge increases, there is a promising opportunity to apply modern molecular technologies under a systems immunology approach to accelerate progress. Furthermore, the limited availability of clinical samples from early clinical trials is a major motivating factor for using advanced technologies to gain the most knowledge from the samples that exist. Broadly speaking, systems immunology encompasses the tools of systems biology framework to the unique biological characteristics of innate and adaptive immunity (10,11). The modern roots of systems biology date to the turn of the 21st century, as genome sequencing technologies began to rapidly accelerate (12,13). Under a systems biology paradigm, biological entities are modeled as a network of interacting units, typically requiring high-throughput data generation paired with integrative computational and statistical models. Each foundational unit may be defined as an intracellular entity such as a gene or metabolite, or a cellular entity such as an individual cell or cell type (10). Technological advances have made possible many approaches that were previously untenable except under theoretical circumstances. Massively parallel molecular assays can quantify not only the levels of various molecular analytes such as RNA, protein, and metabolites, but also interrogate their interactions at increasingly higher throughput. These molecular assays can also be applied to measure the cellular states and interactions after either genetic or chemical perturbations in a high-throughput fashion (14–16). The advent of single-cell technologies in particular has led to an influx of data, and ongoing consortia such as the Human Cell Atlas (17), the Human BioMolecular Atlas Program (18), and the Human Tumor Atlas Network will provide key resources to characterize the underlying basis of normal and malignant cellular phenotypes and tumor microenvironment. The immune system has been an important focus in multiple consortia, leading to opportunities to integrate knowledge into the field of CAR T-cell therapy and other immunotherapies.

In parallel to experimental technology development, a vast array of computational and statistical approaches have been developed to advance systems biology. Mathematical models of tumor-immune interactions have been developed using principles of optimal control theory, treating cellular populations as interacting units within a dynamic system (19,20). Another key class of systems biology algorithms model biological processes as molecular interaction networks, and both computational and technological advancements have allowed for construction and analysis of these networks at a tissue- or cell type- specific resolution (21,22). The relatively high sample sizes produced by single-cell experiments have been increasingly capable of powering for machine learning methods, such as probabilistic graphical models and deep learning, to interpret the data and identify regulatory mechanisms (23–25). A major ongoing challenge in the field is the integration of data across experimental protocols and molecular assays, and approaches such as factor analysis and transfer learning have been developed to address this challenge (26,27).

The mechanisms of CAR T-cell treatment success and failure are likely multimodal and involve tumor, T-cell, and microenvironmental factors (9). From a systems biology perspective, tumor-immune interactions can be viewed as a complex adaptive system that underlies the major clinical challenges, from treatment failures to therapeutic toxicities (Figure 1). In hematological malignancies such as B-cell acute lymphoblastic leukemia (B-ALL), cancer cells may be present in large numbers in peripheral circulation as well as the bone marrow, allowing T-cells to have relatively free access to the tumor throughout the body and in the marrow microenvironment. In contrast, CAR T-cell therapy for solid tumors faces multiple additional barriers including T-cell exclusion, vascular and hypoxic constraints, and immune suppression mediated by additional cell types such as cancer associated fibroblasts, tumor-associated macrophages, and myeloid-derived suppressor cells (28,29). Understanding these factors will be a crucial component of improving existing CAR T-cell therapies and developing new therapies for hematological and solid tumors. As the number of clinical and preclinical trials of CAR T-cell therapies expand, a systems biology approach will become a powerful complement to advance biological discovery (Table 1). Paramount to a systems biology approach is an understanding of the strengths and limitations of each technology, including sources of technical variability and the fundamental role of each molecular analyte in a biological system.

Figure 1.

A systems immunology view of biomedical discovery for CAR T-cell therapy. Immunotherapy requires cellular interactions between tumor and immune populations in multiple physiological compartments including the peripheral circulation, tumor microenvironment, and lymphoid tissues. Addressing the key challenges in CAR T-cell therapy will require an understanding of these interactions among cellular phenotypes and microenvironments. Experimental and computational advances in systems immunology provide the capacity to simultaneously profile multiple cellular populations within each anatomical component, making it possible to investigate the tumor-immune interactions that underlie the major challenges in CAR T-cell therapy. Recent technological advances enable investigators to query genomic, transcriptomic, proteomic, and epigenomic features at a single-cell level. A number of methods for joint profiling of multiple molecular features have been developed, and imaging-based approaches have been developed to resolve spatial context of transcriptomic and proteomic data.

Table 1.

A framework for addressing the major clinical challenges of CAR T-cell therapy using systems immunology approaches^*

Clinical challenge	Systems immunology approach
Biomarker discovery	Profiling of serum, tumor, and immune factors prior to infusion Leveraging pre-existing immunological data from other domains using principles of transfer learning
Pre-treatment effects and identifying combination therapies	Modeling the perturbation of immune cell populations in the tumor and peripheral compartments Development of computational methods that account for gene-gene as well as cell-cell interactions between tumor and immune populations
Understanding T-cell aspects to therapy failure	Joint analysis of endogenous and CAR T-cell populations in circulation and in the tumor microenvironment Characterization of the interactions between T-cells and suppressive cell populations
Expanding CAR T-cell therapy to solid tumors	Joint molecular profiling and imaging of cells in the tumor-immune microenvironment Leveraging data generation by ongoing consortia to understand activating and suppressive immune factors
Antigen escape and novel target identification	Understanding genetic and non-genetic causes of antigen loss Systematic identification of candidate targets in tumor and normal tissues
Addressing major toxicities	Understanding immune cell interactions, particularly interactions between lymphoid and myeloid phenotypes Investigating the interactions between CAR T-cells, cytokines, and organ-specific cell types

^*Completed and ongoing clinical trials have brought to light a number of major clinical challenges for CAR T-cell therapy moving forward: Biomarker discovery, pre-treatment effects and identifying combination therapies, understanding T-cell aspects to therapeutic failure, expanding CAR T-cell therapy to solid tumors, antigen escape and novel target identification, and addressing major toxicities. For each of these major clinical challenges, the tools of systems immunology provide a strategy to better understand the complex interactions between tumor and immune cell populations, providing a path forward for the next generation of cellular therapeutics.

Biomarker discovery: leveraging prior immunological knowledge

The identification of prognostic biomarkers of response to CAR T-cell therapy is of great clinical interest due to the potential to aid in risk stratification and identification of patients most likely to benefit from therapy. There is an urgent need to define T-cell characteristics in the CAR T-cell product that predict key attributes such as proliferation and persistence of CAR T-cells after infusion. A number of cellular and molecular biomarkers have been identified in immune checkpoint blockade: circulating factors such as lactate dehydrogenase (LDH) (30,31); immunological markers such as immune cell composition, T-cell activation pathways, lymphocytic infiltration, and T-cell repertoire (32–35); and tumor markers such as PD-L1 signaling (36) and mutational load (36–38). Some, but not all, of these biomarkers have the potential to generalize to CAR T-cell therapy. Transcriptomic phenotypes of lymphocytic activation and migration may be shared between immunotherapy modalities. On the other hand, biomarkers that assess the diversity in antigen recognition, such as T-cell repertoire and mutational load, may be of lesser significance in CAR T-cell therapy compared to immunotherapies that depend primarily on cytotoxicity through endogenous T-cell receptor signaling. The utility of these biomarkers in combined checkpoint blockade and CAR T-cell therapy, or in future CAR T-cell therapies that may require general T-cell activation and epitope spreading for efficacy, has yet to be assessed in clinical trials.

Circulating factors may play a general role in suppression or activation of adaptive immune responses. Lactic acid has been shown to blunt antitumor responses by T-cell and NK-cells (39), and serum lactate dehydrogenase (LDH) has been shown to correlate with clinical outcomes in immune checkpoint blockade. In CAR T-cell therapy, post-infusion serum cytokine markers have been shown to be predictive of cytokine release syndrome (CRS) (40), the most significant characteristic adverse effect of the therapy; but biomarkers predicting or modulating efficacy in either the product or the patient have not been clearly identified. Transcriptional pathways in T-cell populations, such as T-cell exhaustion, have been associated with resistance to both immune checkpoint blockade and CAR T-cell therapy (41,42). Insights related to tumor evasion of adaptive immunity may be shared between CAR T-cell therapy and other therapeutic modalities. Among the best-known example is expression of the programmed death-ligand 1 (PD-L1) on tumor cells, which interacts with PD-1 on T-cells to suppress T-cell activity. In an in vivo mouse model of pleural mesothelioma, tumor-cell PD-L1 expression inhibited CAR T-cell effector functions (43).

The transfer of knowledge from prior studies of immunology and cancer immunotherapy to CAR T-cell therapy will be key to the systems biology approach. The Immunological Genome Project (44) and more recently, the Human Cell Atlas (17), the Human Tumor Atlas Network, and the Immuno-Oncology Translational Network will provide a wealth of reference data on the molecular physiology of multiple immune populations in normal and malignant environments. In clinical and pre-clinical studies of cancer immunotherapy, particularly in immune checkpoint blockade, there has been an expansion in the use of high-throughput technologies to generate large datasets and computational models of tumor and immune populations. For example, Jiang et al. used publicly available bulk RNA-Seq profiles to derive a predictive signature of T-cell exclusion and dysfunction in anti-PD1 and anti-CTLA4 therapy (45); and Krieg et al. used single-cell mass cytometry to assess cell surface protein phenotypes of peripheral blood mononuclear cells at multiple therapeutic time points (46).

The computational framework for biomarker discovery for CAR T-cell therapy from prior immunological data can be formulated under the principles of transfer learning, which involves the exchange of knowledge from a source domain and task to a target domain and task (47). Under this framework, the source domain consists of data generated from a previous immunological study, and target domain consists of data generated from one or more CAR T-cell studies (Figure 2). Knowledge transfer can occur in the form of model parameters, feature representations, or instances in an integrative model. In the case of immunological data, the most direct methods of knowledge transfer can involve gene signatures, pathway signatures, gene network topology and edge weights, or model coefficients to be used as a statistical prior. Transfer learning has been successfully used for prediction of MHC class I protein binding, clinical and histopathological image analysis, integration of molecular data from genomic and RNAi screens, and integration of single-cell RNA-Seq data (48–52). Immune cell populations are among the most common cells to be studied by high-throughput technologies, and the wealth of publicly available immunological data is an opportunity for the development and deployment of integrative computational methods to be used in studies of CAR T-cell therapy.

Figure 2.

The challenge of integrating data from heterogeneous molecular assays and experimental sources. Prior and ongoing consortia have provided a wealth of data for multiple immune cell populations in isolation and in interaction with each other and tumor cells. Integrative data analysis is a major challenge in the field, and leveraging publicly available immunological data may be a valuable approach for systems immunology model development. The use of unlabeled immune cell data may be framed under an unsupervised transfer learning framework, in which the target task may be dimensionality reduction or clustering of immune cell populations. The use of labeled source data - for example, transcriptomic profiles paired with clinical outcome of another immunotherapy modality – may follow a framework of inductive transfer learning, in which the target task is the prediction of CAR T-cell clinical outcome.

Pre-treatment effects and the search for combination therapy

Patients on clinical trials of CAR T-cell therapy to date have received CAR T-cell therapy subsequent to other treatments, often including multiple rounds of cytotoxic chemotherapy. These prior treatment effects may have a profound impact on the immune cells, tumor cells, and tumor microenvironment involved in CAR T-cell therapy. In studies of other immunotherapies, chemotherapeutic agents have been shown to increase immunogenicity of tumor cells through increased antigenicity and depletion of suppressive immune cell populations (53,54) and increase tumor infiltration by T-cells in solid tumors (55). Combination therapy between certain chemotherapeutic agents and immune checkpoint blockade is the standard of care in some cancers (56). It is unclear to what extent these factors apply to CAR T-cell therapy. Lymphodepleting chemotherapy is routinely given to patients prior to CAR T-cell infusion, and has been associated with improved in vivo expansion and persistence (57), owing to cytokine upregulation as well as suppression of regulatory T-cells and myeloid-derived suppressor cells that has been observed in earlier adoptive cell transfer therapies (58). Clinical and preclinical studies comparing the impact of prior therapy on CAR T-cells themselves are scarce, and such studies are limited by the heterogeneity of prior treatment regimens among patients enrolled in clinical trials. It is possible that certain chemotherapies could have a deleterious impact on CAR T-cell therapy due to their impact on T-cells and activation of pathways hindering expansion in vitro or post-infusion (59).

With over a hundred approved anticancer drugs and over a thousand in development (60), the identification of agents that sensitize cancer cells or augment immunological function in CAR T-cell therapy is a promising objective. Some drug combinations, such as combining immune checkpoint blockade and CAR T-cell therapy, have initial clinical evidence of benefit (61). A highly active area of systems biology has been the development of technologies and computational methods to identify synergistic and antagonistic drug combinations in cancer therapy. High-throughput drug sensitivity screens on cell lines, paired with genomic and transcriptomic data, have provided a resource for modelling individual drug perturbations in cancer cells (62,63), and CRISPR screens of gene pairs provide a functional assay to assess for genetic interactions among known drug targets in cancer cells (64). A number of computational methods have been developed to model the gene regulatory networks of cancer cells, with the objective of identifying synergistic drug combinations or pairs of potentially targetable genes that exhibit synthetic lethality or synthetic dose lethality (65–68). These approaches, however, are challenging to generalize to immunotherapy. Immunotherapy fundamentally differs from conventional pharmacotherapy, in that the mechanism of tumor cell killing is dependent on complex interactions between immune cells and tumor cells, which are not adequately modeled in assays of cancer cell lines. One approach has been to identify differentially expressed genes and gene modules between immunotherapy responders and non-responders, and consider regulators of these pathways as candidate targets. For example, Lesterhuis et al. performed network analysis of immunotherapy responsive and resistant tumor transcriptomes to identify candidate target genes to augment checkpoint blockade (69), and Jerby-Arnon et al. performed single-cell RNA-Seq on 33 melanoma tumors to identify an immune evasion cellular program that suggested CDK4/6 inhibition in combination with immune checkpoint blockade (70). Intratumor targets, however, only represent a subset of drug targets among the potentially targetable cellular populations. For example, stimulation of innate immunity, modification of the tumor immunosuppressive environment, and modulation of T-cell function represent classes of potential drug targets directed toward non-tumoral cells (71). To address these broader opportunities of combination therapies, computational models must consider immune cell populations and tumor-immune interactions.

Use of single-cell technology to better understand the T-cell contribution to CAR T-cell therapy

The underlying T-cell populations used for generation of CAR T-cell products likely play a substantial role in determining the ability of the CAR T-cells to proliferate and kill tumor cells in vivo. Until recently, most protocols for CAR T-cell generation have not selected for T-cell subtypes for engineering, and thus contain a heterogeneous mix of CD4⁺ and CD8⁺ T-cells along the full spectrum of naive to effector phenotypes. A recent exception to this is the JCAR017 approach, using defined populations of CD4 and CD8 cells, with impacts on efficacy and persistence that are still being studied (72). A number of studies have assessed T-cell predictors of therapeutic response in chronic lymphocytic leukemia (42), non-Hodgkin lymphoma (73), and multiple myeloma (74). Fraietta et al. found that sustained response to CAR T-cell therapy in chronic lymphocytic leukemia was associated with increased prevalence of CD27⁺CD45RO⁻CD8⁺ memory-like T-cells prior to CAR T-cell generation, and the presence of CD27⁺PD-1⁻CD8⁺ CAR T-cells expressing high levels of the IL-6 receptor-β chain was predictive of therapeutic response (42). In a study of non-Hodgkin lymphoma, Rossi et al. applied a single-cell secreted cytokine assay on pre-infusion CAR T-cells, and developed a predictive score based on the proportion of cells secreting multiple cytokines and the mean fluorescence intensity of secreted proteins per cell; they reported that their score was significantly associated with treatment response (73). In a study of multiple myeloma, Cohen et al. found that a higher CD4:CD8 ratio in the initial leukapheresis product was associated with greater CAR T-cell expansion in vivo (74). An important area of study has been understanding of the role of early memory phenotypes (75), such as naive T-cells, in establishing proliferative populations and long-term memory. Selection or enrichment of T-cell subtypes has been proposed as a major area of interest for improving CAR T-cell therapies (76,77), which may be aided by the recent proliferation of single-cell transcriptomic, epigenomic, and proteomic technologies.

T-cell subset assessment has conventionally relied on the use of individual cell-surface protein markers, but recent technologies have made it possible to interrogate large numbers of markers at a single-cell level in combination with other molecular and imaging assays. Single-cell mass cytometry uses flow cytometry and mass spectrometry with discrete isotopes to characterize large numbers of cell surface protein markers simultaneously at a single-cell level (78). Simultaneous profiling of single-cell RNA transcriptomes with proteomics has been made possible with CITE-Seq and REAP-Seq, methods that combine single-cell RNA-Seq with oligonucleotide-labeled antibodies that bind to cell surface proteins (79,80). These methods may be used to distinguish between CAR T-cells and non-CAR T-cells in single-cell transcriptomic studies, allowing for the capacity to identify gene expression pathways specific to the endogenous and engineered cell populations.

Single-cell RNA-Seq has been a breakthrough technology for characterizing whole-transcriptome gene expression at a single-cell resolution. Single-cell RNA-Seq has been used to profile infiltrating T-cells in several solid tumor types (81,82), demonstrating the ability to investigate T-cell populations within the tumor microenvironment. Tikhonova et al. and Baryawno et al. recently characterized the mouse bone marrow microenvironment using single-cell RNA-Seq, providing key insights into cellular phenotypes and signaling that underlies the bone marrow niche (83,84). Applying single-cell transcriptomics to complex populations of peripheral and infiltrating T-cells in CAR T-cell therapy may aid in identifying subpopulations and cellular pathways that mediate proliferation, persistence, and memory formation. Characterization of suppressive cell types in the tumor microenvironment, such as tumor-associated macrophages, cancer-associated fibroblasts, myeloid-derived suppressor cells, and regulatory T-cells, may aid in understanding the contribution of these cells in the efficacy of CAR T-cell therapy. Without doubt, current single-cell technologies still have a number of limitations, such as coverage and throughput. There has been significant work to address these and other technical challenges in single-cell RNA-Seq, including variability in sequencing depth and cell size, drop out, and batch effect correction (85).

Improving the engineering of CAR T-cells with the goal of increasing efficacy, speeding manufacturing, and decreasing toxicity is a central challenge in the field, with the potential to enhance existing therapies and accelerate the adoption of cellular therapeutics to other cancer types. Beyond cancer therapy, CAR T-cells have been designed for other therapeutic purposes, including autoimmune disease (86), prevention of allogeneic transplant rejection (87), and against infectious disease (88,89). Modeling of intracellular and extracellular interactions may play an important role in the design of novel engineering of CAR components. The intracellular pathway mediating cytotoxic response in CAR T-cells is incompletely understood, and recent studies have applied computational modeling to understand phosphorylation kinetics of certain CAR constructs (90). A number of synthetic biology approaches have been designed to augment and enhance CAR construction, including T-cells engineered with kill switches, growth switches, chemokine receptors, multiple targets, and avoidance of allogeneic rejection (91,92).

Addressing challenges in solid tumors

CAR T-cell therapies have seen their greatest success in hematological B-cell malignancies, and the design of an effective CAR T-cell therapy for solid tumors is a key objective in pediatric and adult clinical oncology. In contrast to hematological cancers, solid tumors pose particular challenges including the need for CAR T-cell trafficking and infiltration, overcoming a hypoxic and metabolically constrained tumor microenvironment, and the presence of suppressive cell populations (29). These factors have been hypothesized to be major barriers to the effective implementation of CAR T-cell therapy in solid tumors.

In CAR T-cell therapy, T-cells are typically extracted from and re-infused into the peripheral circulation, making it necessary for CAR T-cells to enter the tumor microenvironment. To address this challenge, some groups have explored intracavitary or intratumoral injection of CAR T-cells (93,94). This strategy may be technically challenging, and even more importantly, it is unclear to what extent regionally delivered CAR T-cells can migrate to other tumor sites in highly disseminated disease. The level of T-cell infiltration into solid tumors has been correlated with chemokine expression in the tumor microenvironment, and variability in chemokine expression may explain the low level of T-cell infiltration in some tumors (95). Some groups have focused on engineering approaches to localize CAR T-cells, such as the manufacturing of T-cells with receptors for chemokine CCR2 to increase cellular migration (96) or injection of amphiphile CAR T-cell ligands that traffic with endogenous albumin to lymph nodes and enhance the interaction between CAR T-cells and antigen-presenting cells (97).

Recent technological advances in proteomics and single-cell RNA-sequencing has improved our understanding of the composition and cellular interactions within the tumor-immune microenvironment (98). For example, Spitzer et al. performed mass cytometry of immune cells in multiple anatomical compartments of a mouse model of cancer immunotherapy, and demonstrated that immune activation occurs systemically in initial therapy but only in the peripheral compartment during tumor rejection (99). In particular, the role of regulatory T-cells, M2 macrophages, and myeloid-derived suppressor cells in immune evasion has been demonstrated in clinical and pre-clinical studies of other immunotherapies in solid tumors (100); however, limited sample sizes and technical challenges in raising immunocompetent mouse models have made it challenging to directly study these interactions of CAR T-cells with these cellular populations (28). Ongoing consortia, such as the Human Tumor Atlas Network and the Adult Immunotherapy Network, have dedicated significant resources into generating publicly accessible data that will be key to unravelling the basis of immune suppression in the tumor-immune microenvironment. An exciting advance in the field has been the integration of cellular imaging with single-cell molecular profiling. Technologies such as Spatial Transcriptomics and Slide-Seq have been recently developed to integrate single-cell RNA-Seq with cellular imaging (101,102), providing an opportunity to understand the spatial context of cellular populations in the tumor-immune microenvironment. Highly multiplexed analysis of proteins in the context of tissue imaging has been made possible with several imaging technologies, including Imaging Mass Cytometry (IMC) (103), Multiplexed Ion Beam Imaging (MIBI) (104), and co-detection by indexing (CODEX) (105). These resources will provide rich datasets for modeling the signal transduction pathways within the tumor microenvironment in CAR T-cell therapy.

Identifying tumor-specific antigens is a challenge for the development of future CAR T-cell therapies for non-hematopoietic malignancies. One of the reasons for the success of CAR T-cell therapies for B-cell malignancies is that the ablation of normal B-cell populations is a manageable toxicity via immunoglobulin replacement. Cancer-associated antigens have to be chosen carefully to avoid off-tumor, on-target toxicity, which has been a challenge in some solid tumor CAR T-cell trials (14). Computational and functional screens have been designed to identify cell surface proteins that are ubiquitously expressed on tumor cells but not on normal cells (106,107). A major challenge to this approach is the lack of a comprehensive reference catalog of normal cell populations. Consortia such as GTEx have performed bulk RNA-Seq on multiple normal human tissues (108), but bulk assays may miss rare but important cell populations such as adult stem cells and non-parenchymal cells. Single-cell data sources have the potential to address this problem through their capacity to characterize rare and previously unknown cell populations (109). Designing CAR T-cell therapies to target multiple antigens may be an approach to circumvent antigen escape, as well as increase tumor specificity. Toward this goal, an integration of single-cell atlases with data from individual studies may provide the much-needed reference from which tumor-specific antigens and combinatorial antigen targets may be identified.

Therapy-induced toxicity

CRS is the most common and significant toxicity of CAR T-cell therapy, reported to occur in 54–91% of patients receiving anti-CD19 CAR T-cell therapy (110). This syndrome was not apparent in preclinical models, and only became manifest in early clinical trials (2). Neurotoxicity, referred to as immune effector cell-associated neurotoxicity syndrome, or ICANS (111), has been observed as an adverse event, with the most concerning trait being cerebral edema. While the underlying pathophysiology of these toxicities remain incompletely understood, clinical trials have implicated macrophages, IL-1, endothelial activation, and especially IL-6 in the pathogenesis of CRS. IL-6 blockage with tocilizumab is the only currently approved therapy for CRS (2,112). Novel murine models have been recently developed, and have suggested a role of myeloid lineages and endothelial activation in CRS (113). Giavirdis et al. (114) found that CAR T-cells recruit and activate macrophages, leading the myeloid cells to produce IL-6 and iNOS which contribute CRS. Norelli et al. (115) found that monocytes and macrophages were primarily responsible for increases in IL-1 and IL-6 in their mouse model of CRS, with IL-1 production preceding and possibly stimulating IL-6 production. In a clinical trial of CD19 CAR T-cell therapy in B-ALL, CLL, and NHL, investigators reported evidence of increased endothelial activation in patients with severe CRS, via elevated Von Willebrand Factor and increased Ang2:Ang1 ratio (110). Moreover, IL-6 has been shown to disrupt the blood-brain barrier in vitro, suggesting a possible mechanistic explanation for neurotoxicity that has been reported in some patients.

Toxicities of CAR T-cell therapy such as CRS and ICANS remain as important concerns that may or may not be class effects of the therapy, and prognostic and therapeutic strategies will require a deeper understanding of the underlying biological basis. While initial clinical studies have focused on tumor cells and T-cells, a systems immunology approach will involve investigation of cell-cell interactions with various other immune cell types in the peripheral and marrow microenvironments. Comprehensive analyses of myeloid and lymphoid compartments in clinical trials, as well as expansion of murine models, will be important paths forward for the study of these therapy-limiting toxicities.

Conclusions and future directions

CAR T-cell therapy has been a breakthrough in clinical oncology. Understanding the T-cell, tumor-cell, and tumor microenvironment factors mediating therapeutic efficacy will play a significant role in optimizing current therapies and applying CAR T-cells to other tumor types. Modern omics and systems biology technologies provide opportunities for biomarker discovery and therapeutic development. As clinical cohorts of patients treated with CAR T-cell therapy continue to grow, there will be increasing availability of clinical samples from which rich multi-omic profiling can advance the discovery process. Like many biomedical challenges in the 21st century, translational research will be best performed through a collaborative effort by interdisciplinary teams of clinicians, experimental and computational biologists.