Advancing CAR T cell therapy through the use of multidimensional omics data

Abstract

Despite the notable success of chimeric antigen receptor (CAR) T cell therapies in the treatment of certain haematological malignancies, challenges remain in optimizing CAR designs and cell products, improving response rates, extending the durability of remissions, reducing toxicity and broadening the utility of this therapeutic modality to other cancer types. Data from multidimensional omics analyses, including genomics, epigenomics, transcriptomics, T cell receptor-repertoire profiling, proteomics, metabolomics and/or microbiomics, provide unique opportunities to dissect the complex and dynamic multifactorial phenotypes, processes and responses of CAR T cells as well as to discover novel tumour targets and pathways of resistance. In this Review, we summarize the multidimensional cellular and molecular profiling technologies that have been used to advance our mechanistic understanding of CAR T cell therapies. In addition, we discuss current applications and potential strategies leveraging multi-omics data to identify optimal target antigens and other molecular features that could be exploited to enhance the antitumour activity and minimize the toxicity of CAR T cell therapy. Indeed, fully utilizing multi-omics data will provide new insights into the biology of CAR T cell therapy, further accelerate the development of products with improved efficacy and safety profiles, and enable clinicians to better predict and monitor patient responses.

Introduction

Chimeric antigen receptors (CARs) are synthetic fusion receptors designed to redirect T cells to recognize and eliminate cancer cells that express cognate antigens¹. To achieve this aim, CARs have typically been constructed using antibody-based moieties targeting specific tumour-associated antigens, which are linked via a hinge and transmembrane domain to intracellular signalling motifs derived from both the T cell receptor (TCR) CD3ζ chain and a co-stimulatory receptor (such as CD28 and/or 4–1BB)¹. Transgenes encoding these constructs are introduced into autologous (patient-derived) or allogeneic (healthy donor-derived) T cells ex vivo, followed by selection and in vitro expansion of CAR-expressing T cells before they are infused into the patient. CAR T cell therapy has moved to the forefront of therapy for relapsed and/or refractory B cell malignancies, owing to impressive response rates and durability in patients with B cell acute lymphoblastic leukaemia (B-ALL)², large B cell lymphoma (LBCL)³, mantle cell lymphoma⁴, follicular lymphoma⁵ or multiple myeloma⁶. In addition, CAR T cells have been demonstrated to induce clinical responses in patients with various advanced-stage solid tumours, including glioblastoma⁷, pancreatic ductal adenocarcinoma⁸, sarcoma⁹, gastrointestinal cancers¹⁰ and castration-resistant prostate cancer¹¹, albeit less frequently and more transiently than in patients with B cell malignancies. Nevertheless, the clinical success of CAR T cell therapy faces several challenges, including the paucity of robustly expressed tumour-specific antigens in many malignancies^12,13, CAR T cell exhaustion^14,15, limited CAR T cell persistence^16,17 and potential life-threatening toxicities^18,19.

Multi-omics data, encompassing genomics, epigenomics, transcriptomics, proteomics and beyond, can provide deep insights into the abundance and/or variation of biological molecules across multiple dimensions in various tissues or cells^20,21. The rapid development of multidimensional profiling strategies and approaches for their integration has facilitated the use of multi-omics data to characterize the complex molecular features of diverse cellular processes in broad research fields^22,23, including CAR T cell therapy^24–26. CAR T cells are living cells that can actively sense and respond to a wide variety of extrinsic and intrinsic factors²⁷. Therefore, identification of detrimental phenotype heterogeneity and pathway regulation of pre-manufacturing autologous T cells through bulk and single-cell profiling^28–30 might contribute to the quality control of CAR T cell products or inform strategies for CAR design or product manufacture that improve their functionality. Moreover, the use of multi-omics strategies to monitor the temporally dynamic expansion of functional and/or dysfunctional CAR T cell subpopulations^24,31, characterize CAR T cell transcriptomic³², epigenomic³³ and metabolic³⁴ dynamics, delineate tumour microenvironment (TME) features that restrict CAR T cell infiltration and suppress CAR T cell cytotoxicity³⁵, determine the origins and functional consequences of key cytokines^36,37, and predict on-target, off-tumour toxicities in non-malignant tissues³⁸ could facilitate biomarker discovery and improve our mechanistic understanding of CAR T cell therapy responses and toxicities.

In this Review, we highlight the potential of multi-omics data to advance CAR T cell therapy. We first provide an overview of the multidimensional profiling technologies that have been utilized in the investigation of CAR T cell therapy. Next, we discuss strategies leveraging multi-omics data to discover single and combinatorial CAR targets. We further discuss molecular features and other key factors that are associated with enhanced antitumour responses and/or decreased toxicity of CAR T cell therapy as uncovered by multi-omics profiling. Exploiting current and emerging multidimensional omics profiling technologies will enable a comprehensive understanding of the molecular biology of CAR T cell therapies, ultimately enabling the clinical benefits of these treatments to be maximized.

Multidimensional profiling technologies

Data from genomics, epigenomics, transcriptomics, TCR-repertoire profiling, proteomics, metabolomics and microbiomics studies have been applied to address the remaining key challenges for CAR T cell therapy. Leveraging such multi-omics data can provide novel insights into tumour and non-malignant cell characteristics, cell state transitions, cell types, and cell–cell interactions in the TME that affect treatment outcomes of CAR T cell therapy (Fig. 1 and Table 1).

Fig. 1 |. Overview of applications of multi-omics data in CAR T cell therapy.

Several types of samples, including bulk tissue specimens, single-cell suspensions or spatially preserved tissue slides (left panel), can be used for multidimensional omics analyses, including genomics, epigenomics, transcriptomics, T cell receptor (TCR)-repertoire, proteomics, metabolomics and/or microbiomics profiling (middle left panel), across a multitude of applications in the field of chimeric antigen receptor (CAR) T cell therapy. The valuable multi-omics data facilitates the characterization of diverse tumour and non-malignant cell features, cell functional phenotypes (for example, of CAR T cells both prior to and after infusion), and various properties of the tumour microenvironment (middle right panel), which will enable advances in the safety and efficacy of CAR T cell therapy through improvements in target identification and cell engineering (right panel).

Table 1 |.

Representative multi-omics profiling approaches applied to study CAR T cell therapy

Omics approach	Profiling technology	Examples of insights obtained
Genomics	DNA-seq	Identification of genomics alterations as potential targets or as biomarkers for therapeutic outcome44–46
	CRISPR screening	Defining dependencies for the cytotoxic activity of CAR T cells and determinants of the sensitivity of cancer cells to CAR T cell-mediated cytotoxicity52,53
Epigenomics	DNA methylation arrays	Definition of differentially methylated genomic loci related to CAR T cell functions and therapeutic responses33,66
	ATAC-seq	Characterization of changes in accessible chromatin regions of genes that affect CAR T cell functions32,68
	ChIP-seq	Identification of transcriptional reprogramming of the exhaustion-associated epigenome in exhausted CAR T cells63
	scATAC-seq	Discovery of epigenetic differences correlated with distinct CAR T cell subpopulations28,71
Transcriptomics	RNA-seq	Identification of transcriptional changes related to CAR T cell functions32,76; discovery of novel targets based on differential expression and RNA dysregulation in malignant versus non-malignant tissues80,82
	scRNA-seq	Characterization of CAR T cell functional states that are associated with particular cellular conditions, CAR designs/cell products, clinical responses and toxicities87; discovery of subpopulations of non-malignant and tumour cells that cause toxicities and disease relapse, respectively90,203; identification of hostile factors in the tumour microenvironment that affect CAR T cell function and, therefore, hamper solid tumour elimination91
TCR-repertoire profiling	TCR-seq	Kinetic profiling of different TCR clonotypes throughout the treatment process96,97
Proteomics	Mass spectrometry	Information on the surfaceome (expression landscape of cell-surface proteins) and immunopeptidome (repertoire of MHC-presented peptide epitopes) as a resource for CAR target identification100,103,106; phosphoproteomics characterization to evaluate CAR downstream signalling109
	Mass cytometry	Identification of immune markers for monitoring of CAR T cell activation, proliferation and exhaustion114
Metabolomics	Mass spectrometry	Discovery of metabolic reprogramming in CAR T cells, which could enhance antitumour efficacy34,124,125
	Metabolic flux analysis	Assessment of dynamic metabolic pathway activity related to different cell conditions129
	Mass cytometry of metabolic proteins	Characterization of the metabolic adaptation of CAR T cell products after infusion242
Microbiomics	16S rRNA-seq and metagenomics shotgun sequencing	Identification of features of the gut microbiota correlated with response and toxicities141

16S rRNA-seq, 16S ribosomal RNA sequencing; ATAC-seq, assay for transposase-accessible chromatin with high-throughput sequencing; CAR, chimeric antigen receptor; ChIP-seq, chromatin immunoprecipitation sequencing; DNA-seq, DNA sequencing; RNA-seq, RNA sequencing; scATAC-seq, single-cell ATAC-seq; scRNA-seq, single-cell RNA-seq; TCR, T cell receptor; TCR-seq, TCR sequencing.

Genomics

Comparisons of DNA sequencing (DNA-seq) data from tumour and non-malignant tissue samples have identified numerous tumour-associated somatic mutations, some of which might generate tumour-specific neoantigens that are presented on the cell surface and could potentially serve as novel targets for CARs or TCR-like constructs^39,40. To date, however, attempts to target cancer cells with neoantigen-directed CAR T cells have been limited to a few preclinical and clinical studies^41–43. Alternatively, tumour-associated mutations and complex genomics alterations might influence clinical responses to CAR T cell therapy^44–47. Genome-wide pooled CRISPR–Cas9 knockout library screening followed by sequencing-based quantification of the frequencies of single guide RNAs enables large-scale, high-throughput studies to assess the influence of individual genes under different conditions and thereby identify cellular functional dependencies^48,49. Such functional genomics profiling has resulted in the identification of key genes involved in T cell activation⁵⁰, persistence⁵¹, cytotoxicity^52–54 and dysfunction⁵⁵ as well as genetic alterations in tumour cells that influence resistance to treatment⁵⁶, highlighting regulators that could potentially be exploited to enhance the efficacy of CAR T cell therapy. Meanwhile, pooled CRISPR-based gain-of-function screening in combination with single-cell RNA sequencing (scRNA-seq) has been used to measure the abundance and assess the cellular states of T cells expressing different knock-in constructs and highlighted a novel chimeric cytokine–co-stimulatory receptor for CAR T cells that improves solid tumour clearance⁵⁷. In summary, functional genomics profiling is a powerful method to elucidate mechanistic regulators of CAR T cell functions that might ultimately be leveraged to refine CAR design or cellular engineering.

Epigenomics

Epigenetic factors, including DNA methylation, histone modifications, and chromatin accessibility and 3D architecture, enable the integration of multiple cellular signals to dynamically regulate the gene expression programme and, ultimately, the phenotype and function of T cells^58,59. Epigenetic reprogramming of CAR T cells through perturbation of epigenetic modulators therefore has the potential to boost antitumour efficacy^60–64. Genome-wide epigenomics profiling techniques have become the optimal approach for investigating epigenetic regulatory landscapes⁶⁵, including those of CAR T cells. For example, DNA methylation array analyses can help to identify differentially methylated sites associated with response to CAR T cell therapy^33,66. Combination analyses with assay for transposase-accessible chromatin using sequencing (ATAC-seq) and RNA sequencing (RNA-seq) can facilitate the identification of epigenomics regulatory modules that control cellular programmes related to CAR T cell exhaustion and cytotoxicity^32,66–69, thus providing a multitude of insights into potential epigenetic reprogramming strategies to improve efficacy.

Advances in single-cell epigenomics profiling⁷⁰ have enabled evaluation of the epigenetic regulatory landscape within individual CAR T cells⁷¹. Combining data derived using single-cell ATAC-seq with transcriptomics and proteomics data has helped to identify the patterns of epigenetic regulation associated with functional subpopulations of CAR T cells²⁸. Collectively, the emerging evidence from epigenomics profiling studies demonstrates the importance of dynamic epigenetic alterations within CAR T cells in determining the molecular features associated with therapeutic efficacy.

Transcriptomics

RNA-seq has an essential role in understanding cellular states and has enabled tremendous progress in the characterization of tumour heterogeneity and the TME^72–74. In the field of CAR T cell therapy, RNA-seq is most commonly applied for differentially expressed gene analysis under various conditions to delineate the cell-state transitions that are associated with efficacy^15,32,75,76. Moreover, monitoring changes in the transcription of genes encoding various cytokines, which are key mediators of CAR T cell activity, is also crucial to improve efficacy and reduce the risk of cytokine-release syndrome (CRS), immune effector cell-associated neurotoxicity syndrome (ICANS) and other toxicities^67,77–79. Application of total RNA-seq provides additional information, for example, on the expression of non-coding RNAs that can regulate immunomodulatory molecules and thereby affect CAR T cell cytotoxicity⁸⁰. Furthermore, dysregulation of post-transcriptional RNA processing can lead to variant transcripts, thus potentially generating different protein isoforms that enable escape from CAR recognition⁸¹. Therefore, increasing interest is being placed on exploiting the existing and expanding tumour RNA-seq data to discover novel CAR targets derived from the differential expression and dysregulation of RNAs^82,83.

scRNA-seq, which generates data on the transcriptome of individual cells, has become a well-established method since 2009 (ref. ⁸⁴) and is widely applied in cancer research^85,86. Notably, scRNA-seq is an emerging method of investigating features of CAR T cells in preclinical and translational research. For example, scRNA-seq enables gene expression-based cell state examination of CAR T cell infusion products, with data demonstrating that signatures of cytotoxicity, central memory, exhaustion and other cellular programmes can be affected by variations in CAR T cell conditions or designs^52,76,87,88 and are associated with clinical response and toxicity¹⁵. Beyond profiling of the CAR T cell products, applying scRNA-seq to non-malignant tissues and tumours can help to identify rare cell subpopulations in order to elucidate the mechanisms underlying treatment-related toxicities^15,89,90. Additionally, scRNA-seq can provide rich gene expression data across various cell types and is therefore an optimal unbiased approach for exploring factors in the TME that might affect the antitumour activity of CAR T cells^91,92. Such insights could offer potential opportunities to overcome immunosuppressive features of the TME and seek solutions for the expansion of CAR T cell applications in solid tumours^91,93.

TCR-repertoire profiling

The TCR is a heterodimeric cell-surface protein complex comprising heterogeneous TCRα and TCRβ or TCRγ and TCRδ chains generated through random genetic recombination of different variable, diversity, junctional and constant gene segments in individual T cell clones; this process results in a varied T cell population with a highly diverse TCR repertoire to mediate antigen-specific adaptive immune responses⁹⁴. High-throughput technologies have been developed to sequence the TCR chains expressed in each T cell and thus to characterize the TCR repertoire and clonotypic diversity of the T cell population. Performing TCR sequencing in combination with other omics approaches, such as transcriptomics profiling, enables specific TCR clonotypes to be linked with varying T cell phenotypes⁹⁵. This combined analysis can help to characterize T cell dynamics through tracing of T cell clonality among different subpopulations over time⁹⁵, and such studies have demonstrated the cytotoxic and proliferative features of highly expanded CAR T cell clonotypes in patients with cancer^96,97. Moreover, advances in single-cell technologies have enabled simultaneous TCR sequencing and scRNA-seq in individual cells, which assigns a specific TCR clonotype to an individual T cell with a distinct transcriptional phenotype and could thus enable TCR clonotypes to be used as a surrogate for the expansion and persistence of T cell functional states throughout therapy¹⁵. With the caveat that individual TCR clonotypes can encompass a mixture of cell states along a functional continuum, these profiling methods offer a route to further explore whether specific clonotypically defined T cell states are associated with CAR T cell performance or toxicities.

Proteomics

Proteomics profiling enables the identification and quantification of proteins and their post-translational modifications (such as phosphorylation and glycosylation) to characterize cellular pathway activity and cell functions. For example, emerging proteomics profiling approaches in the field of immuno-oncology have substantially improved our understanding of the mechanisms of tumour immune evasion⁹⁸. Notably, applying single-cell proteomics profiling in the context of cancer immunotherapy has shown early success in uncovering the clinical relevance of distinct immune cell phenotypes in the TME⁹³.

Traditional proteomics profiling technologies, such as protein microarrays that provide data on the expression of select proteins, have been applied to quantify plasma cytokine levels in patients with lymphoma or leukaemia following treatment with CAR T cells⁹⁹. Advances in mass spectrometry-based proteomics techniques have enabled robust high-throughput profiling of the membrane proteome, including the cell surfaceome¹⁰⁰ and immunopeptidome¹⁰¹, as well as signalling pathway dynamics reflected in the phosphoproteome¹⁰². The tumour surfaceome has been a source of novel therapeutic targets in various haematological and solid malignancies^103–105. The immunopeptidome and surfaceome of non-malignant tissues might serve as a reference database for the selection of tumour-specific targets and thus avoid on-target, off-tumour toxicities^100,106. Phosphoproteomics can be utilized to characterize CAR downstream signalling activity to evaluate the functional states associated with different CAR constructs^107–109. Thus, proteomics profiling has deepened our knowledge of various aspects of CAR T cell therapy, especially CAR target discovery and CAR function¹¹⁰.

Flow cytometry is the gold standard approach for both membrane and intracellular protein detection at the single-cell level for immune cell classification¹¹¹. Advances in full-spectrum flow cytometry, whereby highly sensitive detector arrays are used to capture the full emission spectrum of a diverse range of fluorescent molecules, have improved throughput by increasing the number of cellular parameters that can be evaluated in an experimental run, thus presenting a powerful approach for protein expression analysis in studies of immunotherapy¹¹². Mass cytometry, or cytometry by time of flight (CyTOF), combines flow cytometry and mass spectrometry techniques to enable high-throughput, single-cell proteomics profiling (through simultaneous evaluation of >40 proteins in millions of cells during one run)¹¹³ for high-dimensional cellular phenotyping. Mass cytometry analysis of protein expression has been utilized for functional characterization of CAR T cells in preclinical and clinical settings, facilitating rapid monitoring of the dynamic features of these cells^114–116.

Technologies enabling single-cell secretome analyses have also been developed. For example, single-cell microchamber, proteomics barcode and multiplex immunoassay chips have been used to quantify the secretion of ~32 selected effector and immunostimulatory molecules by pre-infusion CAR T cell products as a measure of their ‘polyfunctionality’, which is associated with clinical response and toxicities^117,118.

Metabolomics

Metabolomics profiling catalogues a suite of small molecules produced during metabolic processes, providing a functional readout of various cellular activities¹¹⁹. T cell function is regulated by intrinsic and extrinsic metabolic factors, and robust intrinsic metabolic activity is necessary for effective T cell cytotoxicity¹²⁰. Disrupted metabolism in the TME, for example, owing to hypoxia and/or decreased nutrient availability, can lead to unmet metabolic demands in infiltrating T cells and thus limit their antitumour activity¹⁶. Therefore, maintaining the metabolic fitness of CAR T cells is important to improve their therapeutic efficacy, and especially to enhance their function in the immunosuppressive TME of solid tumours^121–123.

Mass spectrometry-based metabolomics enables both targeted metabolite quantification and comprehensive untargeted metabolite profiling. Several studies have profiled metabolites and their intermediates generated through various biochemical processes of interest in T cells. Targeted quantification of metabolites involved in glycolysis and the tricarboxylic acid cycle has highlighted impaired intrinsic metabolic processes that affect the proliferation and exhaustion of tumour-infiltrating T cells, presenting opportunities to enhance CAR T cell efficacy through metabolic reprogramming^124,125. Metabolomics data can also be leveraged for metabolic reprogramming of CAR T cells to enhance their fitness and resistance to unfavourable extrinsic, microenvironmental factors^34,121,126.

Cell metabolism is variable and adaptable to environmental constraints. Fluxomics describes the production and consumption rates of metabolites involved in intracellular metabolic networks, which reflect the dynamic metabolic processes within the cell¹²⁷. A combination of stable isotopes and mass spectrometry can be used to trace and quantify dynamic metabolic fluxes, referred to as metabolic flux analysis, thereby enabling fluxomics profiling of metabolic pathways of interest such as glycolytic and mitochondrial respiration¹²⁸. This approach has advantages for the dynamic assessment of metabolic processes in CAR T cells with differing CAR designs or under different environmental conditions, with emerging fluxomics data emphasizing the pivotal role of maintaining the metabolic fitness of CAR T cells for effective treatment^123,129,130.

Microbiomics

The microbiota is emerging as a key factor regulating cancer progression and therapeutic responses^131,132. Several studies have demonstrated correlations between the gut microbiome or dietary interventions and clinical responses to immunotherapies, in particular, immune-checkpoint inhibitors targeting CTLA4, PD-1 or PD-L1 (refs. ^133–136). Moreover, microbial intervention through faecal microbial transplantation of favourable microbiota from healthy donors or responders can improve the efficacy of immune-checkpoint inhibitors^134,137.

Two main methods have been developed for the profiling of microbiota: 16S ribosomal RNA-seq (16S rRNA-seq)¹³⁸ and metagenomics shotgun sequencing^139,140. 16S rRNA-seq is more cost-effective for large-scale analyses of microbial taxonomics composition but does not provide information on viral or eukaryotic components¹³⁸. Metagenomics shotgun sequencing comprehensively covers all microorganisms present in the sample, increasing the potential to discover novel species with potential roles in cancer, and provides data at gene-level resolution for analyses of functional variations, although these benefits come at the cost of computational challenges. Microbiomics profiling of faecal samples from patients receiving CAR T cells has revealed correlations between gut microbiota composition and response to therapy¹⁴¹, suggesting promising opportunities to exploit links with the microbiota to improve CAR T cell therapy. The current studies have mainly focused on the effects of the gut microbiota, yet the intratumour microbiota might also have an important influence on the antitumour immune response^132,142, highlighting an additional dimension for future exploration.

Incorporation of spatial contexture into omics profiling

Advances in spatial molecular profiling methods, especially spatial transcriptomics and proteomics, have enabled the large-scale and high-resolution characterization of localized variations in gene and protein expression using tissue sections¹⁴³. Spatial transcriptomics is an emerging technology that enables high-plex or transcriptome-wide quantification of RNAs at near-cellular¹⁴⁴ or subcellular¹⁴⁵ resolution that, when integrated with non-spatially resolved single-cell transcriptomics, might reinforce analyses of cell–cell interactions through the precise location of cells within the tissue architecture¹⁴⁶. Similar approaches are being developed for spatial epigenomics profiling, such as spatial ATAC-seq and spatial cleavage under targets and tagmentation (CUT&Tag-seq), which has proven utility in characterizing epigenetic regulation in tissues at high resolution^147,148. Established technologies for spatial proteomics, such as imaging mass cytometry, enable the simultaneous quantification and localization of ~40 proteins on a tissue slide, with preservation of the tissue architecture¹⁴⁹, to deconstruct the cellular composition of the TME and explore the cytokine milieu at subcellular resolution across various cancer types^150–153. Furthermore, to explore the spatial organization of T cell metabolic programmes in the TME, a study has utilized CyTOF to quantify the expression of proteins involved in the metabolic regulome in an individual cell and coupled CyTOF to multiplexed ion beam imaging by time of flight (MIBI-TOF) to characterize the distribution and expression of these proteins in tissue¹⁵⁴. Spatial fluxomics has also been developed for dynamic metabolic profiling for cancer research¹⁵⁵. These approaches have not yet been applied to CAR T cell research, but the incorporation of spatial contexture into omics profiling holds great promise to increase our understanding of how cell–cell interactions and cellular neighbourhoods influence crucial cellular programmes related to response and toxicity.

Challenges in multi-omics data processing

Despite the broad applications and remarkable progress of multi-omics profiling in the realm of CAR T cell therapy, notable challenges remain with regard to data analysis and interpretation. First, multi-omics technologies using bulk tissues have broad utility in large-scale data generation, but the potential masking of positive signals and averaging of signals in the context of mixed cell populations might reduce the potential for novel discoveries and the reproducibility of results. This concern could potentially be addressed using newly developed deconvolution algorithms, provided that the underlying reference signatures accurately reflect the transcriptional profiles of each cell type and pathological states within the relevant microenvironment¹⁵⁶. Moreover, the revolutionary applications of high-resolution single-cell and/or spatial multi-omics profiling technologies have provided a variety of rich data on CAR T cell therapy, although these bring a level of computational complexity that requires the development and optimization of innovative bioinformatic tools.

Second, appropriately adapting analytical methods to new applications is a challenging yet crucial step towards achieving robust and reproducible results for biological interpretation; careful evaluation and correction for batch effects caused by differences in sample-processing protocols and data-generating platforms is required for most types of omics data¹⁵⁷. Additionally, for single-cell and spatial omics data, sparsity of data and background noise are common challenges that necessitate the use of appropriate computational algorithms for signal detection^143,158.

Third, the kinetics of molecular fluctuations in cells might require further omics studies with a temporally resolved design¹⁵⁹. Indeed, several studies have demonstrated the importance of monitoring multi-omics variations at different stages of CAR T cell therapy to better profile treatment-related biological changes^96,97,160.

Finally, with the accumulating volume of multi-omics data, optimized computational approaches to integrate data from multiple omics modalities are desperately needed. Pioneering studies have demonstrated the ability of a multi-omics machine learning model to predict treatment response¹⁶¹, suggesting the promise of artificial intelligence algorithms for data integration in future studies; however, challenges relating to the complicated analytical procedures and to the biological translation of results remain to be addressed^162–164.

Target discovery for CAR T cell therapy

Identifying CAR targets with high levels of tumour specificity and coverage is crucial to ensure the antitumour efficacy and reduce the toxicity of CAR T cell therapy^16,165,166. In general, the ideal CAR target should be specifically and highly expressed on the surfaces of all tumour cells but not on essential non-malignant cells. However, owing to shared antigen expression between tumour cells and non-malignant cells as well as highly heterogeneous expression among tumour cells, the identification of suitable CAR T cell targets has been challenging, particularly for solid tumours¹³. In this context, the utilization and integration of multidimensional omics data is a promising approach to identifying optimal CAR targets.

Discovery of single CAR targets

Transcriptomics and/or proteomics data from analyses of both malignant and non-malignant tissues have been widely applied to discover CAR T cell targets that are overexpressed in tumours (Fig. 2). In a study using a novel RNA-seq-based pipeline, glypican 2 was identified as a target selectively expressed at high levels on the surface of neuroblastoma cells without being appreciably detected in non-malignant tissues¹⁶⁷. An analysis of data from non-malignant tissues included in the Human Protein Atlas and the Genotype-Tissue Expression (GTEx) project and from tumour samples included in the Human Protein Atlas pathology database and the database for Differentially Expressed Proteins in Cancer, encompassing 78 different tissues, 124 cell types and 20 cancer types, revealed the expression landscape of >100 candidate CAR targets³⁸. In another study, genomics, transcriptomics and immunopeptidomics data were integrated to identify MHC-binding peptides derived from intracellular oncoproteins highly expressed in neuroblastomas; a non-mutated peptide (QYNPIRTTF) encoded by PHOX2B and presented by HLA-A*24:02 was selected as a lead target for the generation of highly specific peptide-centric CARs, which induced complete tumour regression in xenograft models¹⁶⁸. Thus, bulk transcriptomics and/or proteomics profiling has provided large and comprehensive public resources supporting genome-wide screening for CAR targets across diverse tissue, cell and cancer types. However, averaging gene or protein expression data derived from bulk tissues might mask signals from rare cell populations, with potentially important implications for the efficacy and toxicity of CAR T cell therapy.

Fig. 2 |. Applications of multi-omics data in CAR target identification.

Comparative analyses of genomics, transcriptomics and/or proteomics data from tumour and non-malignant tissue samples can facilitate the discovery of novel targets and optimization of targeting strategies for chimeric antigen receptor (CAR) T cell therapies. The top panels illustrate CAR T cell targeting approaches predicated on single targets, including antigens overexpressed on tumour cells and tumour cell-specific neoantigens or MHC-presented neoepitopes; the bottom panels depict strategies based on combinatorial CAR targets using “AND” or “AND-NOT” logic gates.

Single-cell technologies, such as scRNA-seq, enable omics profiling at unprecedented resolution for CAR target discovery, including the identification of potential safety signals. For example, an analysis of three independent large-scale scRNA-seq datasets derived from human brain tissue samples identified a rare population of mural cells expressing CD19, which might contribute to on-target, off-tumour neurotoxicity from CD19-directed CAR T cell therapy⁸⁹. Similarly, an analysis of two extensive scRNA-seq datasets to predict the on-target, off-tumour toxicity landscape of 591 different CAR targets across a wide range of tissue and cell types identified many as ‘potentially risky genes’ (including EGFR)⁹⁰; a data portal, CARTSC, was also made available as a resource providing data on CAR target expression at the single-cell level.

Neoantigens are novel tumour-specific proteins, generated mainly by somatic mutations in the tumour cells, which might constitute ideal targets for immunotherapy^39,40. To screen for neoantigens, comparative whole-exome or whole-genome sequencing of matched tumour and non-malignant samples is usually performed to identify tumour-specific non-synonymous somatic mutations, followed by RNA-seq and mass spectrometry to confirm expression of the mutated mRNAs and peptides¹⁶⁹. Several neoantigens have already been adopted as promising CAR targets for the treatment of haematological or solid malignancies^41,43. For example, EGFR variant III (EGFRvIII) is a neoantigen specifically expressed in certain solid tumours, including a subset of glioblastomas, leading to ongoing clinical trials of EGFRvIII-directed CAR T cells (such as NCT03726515, NCT03283631 and NCT03638206). To date, high-throughput multidimensional omics approaches have enabled the identification of a large number of neoantigens^169–172, providing a rich source of tumour-specific targets that could be further developed for CAR T cell therapy.

Identification of combinatorial CAR targets

Adequately discriminating between tumour and non-malignant cells using a single CAR target has proved challenging owing not only to on-target, off-tumour toxicities but also to antigen escape that results in therapy resistance^{38,100,104,173}. This issue can potentially be overcome by combining multiple CAR targets using Boolean “AND” and/or “AND-NOT” logic gating to increase tumour-targeting specificity and thus reduce toxicity^174–178 (Fig. 2). In brief, CAR T cells with “AND” logic gates can be triggered only if both target antigens are simultaneously expressed on the tumour cells, whereas CAR T cells with “AND-NOT” logic gates can be activated only in the presence of a target antigen expressed on tumour cells and the absence of an antigen normally expressed on non-malignant cells. Multi-omics data integration has demonstrated the potential to improve CAR T cell design through the identification of combinatorial CAR targets based on expression profiles^{38,100,104,173}. For example, an integrated analysis of extensive transcriptomics and proteomics datasets derived from non-malignant and malignant tissues was performed to propose optimal combinatorial CAR targets in acute myeloid leukaemia, with four possible “AND” pairs fulfilling the stringent efficacy and safety criteria¹⁰⁴. A comparative transcriptomics analysis of 3,567 genes encoding surface proteins, based on mRNA expression data derived from tumours included in The Cancer Genome Atlas (TCGA) and non-malignant tissues included in GTEx, identified 179 “AND” and 443 “AND-NOT” pairs of CAR targets¹⁰⁰. Other studies have similarly applied multi-omics data to discover dual targets for “AND” or “AND-NOT” logic-gate CARs^38,173. This strategy can also be applied to identify more complicated combinations of CAR targets such as triple targets. For example, large-scale mRNA expression data on 33 tumour types from TCGA and 34 non-malignant tissues from GTEx have been leveraged in a comprehensive computational screen of 2,358 predicted surface genes to investigate potential combinations of two or three target antigens for “AND” or “AND-NOT” logic-gated CARs¹⁷³. This study demonstrated the better discriminatory power of combinatorial CAR targets compared with single targets¹⁷³; data on the performance of all predicted antigen pairs are available through the Antigen Explorer portal. Although Boolean logic-gated CARs have not yet been used within the clinical setting, other combinatorial antigen-targeting strategies that mitigate the potential for antigen escape, for example, those predicated on the expression of CLL1 and CD33 in acute myeloid leukaemia^179,180, CD19 and CD22 in B-ALL and LBCL¹⁸¹, and BCMA and GPRC5D in multiple myeloma¹⁸², have shown promise in enhancing the antitumour activity of CAR T cells in preclinical and/or clinical studies.

Single-cell expression data also provide high resolution for predicting the efficacy and safety of CAR T cells with combinatorial targets. For example, an analysis⁹⁰ using scRNA-seq data has predicted potential toxicity risk for a predefined list of CAR target pairs identified based on bulk expression data³⁸. Theoretically, full utilization of comprehensive single-cell multi-omics data from both tumour and non-malignant tissues would be required to nominate optimal target antigens for CAR T cell therapies and thus limit the risk of toxicities. Several scRNA-seq data resources provide a human non-malignant cell atlas encompassing multiple organs and cell types^183–188, and large-scale, pan-cancer, single-cell atlases are in development such as those being mapped by the Human Tumour Atlas Network¹⁸⁹. However, single-cell technologies also have limitations, including the large fraction of zero values of scRNA-seq measurements, which could reflect either transcript drop-out or a true absence of gene expression. At present, the discovery of optimal CAR T cell targets with both high efficacy and safety might best and most feasibly be accomplished by first nominating promising targets that are highly and ubiquitously expressed in tumour cells based on bulk omics data, then using single-cell omics data to predict the risk of on-target, off-tumour toxicities for the nominated targets. Notably, identification of CAR targets based on RNA-seq data is not straightforward, requiring integration with data on the human surfaceome¹⁹⁰ and validation of target expression through proteomics techniques (such as flow cytometry). Furthermore, expression of the target protein on the cell surface does not necessarily correlate with clinical responses, for example, owing to highly heterogeneous antigen expression in tumour cells¹⁹¹, warranting careful evaluation of CAR T cell products with novel targets in rigorous preclinical studies and early phase clinical trials.

Understanding and enhancing CAR T cell efficacy and persistence

A comprehensive understanding of the biological factors that affect the antitumour efficacy of CAR T cells is crucial to the development of strategies to achieve robust and durable therapeutic responses and maximize the clinical benefit. The feasibility of integrating bulk and/or single-cell multidimensional omics data to study the key features related to treatment efficacy has been clearly demonstrated. The main determinants of the efficacy and persistence of CAR T cells can be classified into four broad categories: T cell states and phenotypes, tumour cell characteristics, the TME and the microbiota (Fig. 3 and Table 2).

Fig. 3 |. Multidimensional omics characterization of molecular features associated with the efficacy of CAR T cell therapy.

a, T cell states and phenotypes, including various features of both pre-manufacture T cells and chimeric antigen receptor (CAR) T cell products, can be evaluated across multiple dimensions using omics technologies. b, Multidimensional omics data can also provide insights on tumour cell characteristics associated with response or resistance to CAR T cell therapy, classified into antigen-dependent and antigen-independent factors. c, Transcriptomics and proteomics data can reveal crucial features of the hostile tumour microenvironment that can be leveraged to enhance CAR T cell efficacy, for example, through interventions to increase CAR T cell infiltration and/or resistance to immunosuppression. d, Microbiomics and metabolomics data can be used to identify potential associations between CAR T cell efficacy and the gut microbiota and/or microbial metabolites. FADD, Fas-associated death domain protein; OXPHOS, oxidative phosphorylation; TCR, T cell receptor; TF, transcription factor; T_reg, regulatory T.

Table 2 |.

Examples of cellular features associated with CAR T cell efficacy identified using multi-omics data

Characteristic	CAR T cell targets	Cancer types	Data types	Key findings	Ref.
Pre-manufacture T cells
Heterogeneity	CD19	B-ALL and Hodgkin lymphoma in patients	RNA-seq, scRNA-seq, CITE-seq and scATAC-seq	A higher proportion of naive and early memory T cells was correlated with longer CAR T cell persistence and greater efficacy	28
CAR T cells
Composition	CD19	LBCL in patients	scRNA-seq, scTCR-seq and DNA-seq	Memory signature of CD8+ T cells in the infusion product was enriched in patients with a complete response; patients with a partial response or progressive disease had CD8+ T cells enriched for an exhaustion signature and with low TCR clonotypic diversity	15
	CD19	CLL in patients	Mass cytometry, CITE-seq, TCR-seq, scRNA-seq and scTCR-seq	CD8+ CAR T cells were prominent at the initial time point of therapeutic response; cytotoxic characteristics of CD4+ CAR T cells were key determinants of long-term remission and CAR T cell persistence	97
Expression of specific genes	CD19	Melanoma, thymoma and colon adenocarcinoma in mouse models	ATAC-seq, RNA-seq and scRNA-seq	Knocking out the three NR4A family transcription factors in CAR T cells enhanced their antitumour activity	29
Epigenetic signature	CD19	B-ALL and NHL in patients	DNA methylation microarray	A specific DNA methylation signature (EPICART) was associated with favourable clinical outcomes	33
Metabolic state	CD22, BCMA or HER2	Leukaemia, myeloma and breast cancer in mouse models	CRISPR gain-of-function screen, RNA-seq and mass cytometry	CAR T cells overexpressing PRODH2 had enhanced metabolic and immune function as well as antitumour activity	34
CAR integration site	CD19	CLL in patients	RNA-seq, DNA-seq and ATAC-seq	Integration of CAR sequence into the genome directly disrupted the expression of TET2, which led to an enhanced antitumour response	62
Intracellular signalling domain	CD19 or mesothelin	B cell lymphoma and pancreatic cancer in mouse models	Mass spectrometry	T cells expressing a CD28-CD3ζ CAR including an additional CD3ε signalling module had reduced cytokine release and enhanced antitumour activity and persistence	109
Tumour cells
Antigen-dependent resistance	CD19	B-ALL in patients	DNA-seq and RNA-seq	CD19 mutations or mis-splicing result in loss of the CAR target antigen from the cell surface and disease relapse	81
	CD19	B-ALL in patients	RNA-seq and ChIP-seq	Lineage switching of B-ALL to a CD19-negative myeloid phenotype resulted in disease relapse	205
Antigen-independent resistance	CD19	B-ALL in patients	Genome-wide loss-of-function screen, ATAC-seq, RNA-seq and scRNA-seq	Knocking out genes involved in pro-apoptotic death receptor signalling in tumour cells reduced their susceptibility to CAR T cell-mediated cytotoxicity	55
	CD19	LBCL in patients	DNA-seq and RNA-seq	TP53 alterations (mutations and/or copy-number loss) in tumour cells were associated with poor complete response rates and unfavourable overall survival	44
Tumour microenvironment
To increase CAR T cell infiltration	EGFRvIII	Glioblastoma in mouse models	RNA-seq	Inhibition of PAK4 improved T cell infiltration and sensitized glioblastomas to CAR T cell therapy	35
To strengthen CAR T cells against immune suppression	CD19 and MUC16	B-ALL and ovarian cancer in mouse models	Mass cytometry	IL-18-secreting CAR T cells could activate endogenous immune cells and modulate the tumour microenvironment, thereby promoting the antitumour immune response	222
Microbiota
Diversity and composition	CD19	B-ALL and NHL in patients	16S rRNA-seq and metagenomics shotgun sequencing	Baseline α-diversity was lower in CAR T cell recipients than in individuals without cancer; the abundance of specific bacterial taxa of Lachnospiraceae, Ruminococcaceae and Bacteroidaceae was higher in patients who had a complete response	141
Microbial metabolites	ROR1	Pancreatic cancer in mouse models	Mass spectrometry	The microbial metabolites pentanoate and butyrate could enhance the antitumour activity of CAR T cells	223

16S rRNA-seq, 16S ribosomal RNA sequencing; ATAC-seq, assay for transposase-accessible chromatin with high-throughput sequencing; B-ALL, B cell acute Lymphocytic Leukaemia; CAR, chimeric antigen receptor; ChIP-sequencing, chromatin immunoprecipitation sequencing; CITE-seq, cellular indexing of transcriptomes and epitopes by sequencing; CLL, chronic lymphocytic leukaemia; DNA-seq, DNA sequencing; EGFRvIII, EGFR variant III; LBCL, large B cell lymphoma; NHL, non-Hodgkin lymphoma; RNA-seq, RNA sequencing; scATAC-seq, single-cell ATAC-seq; scRNA-seq, single-cell RNA-seq; scTCR-seq, single-cell T cell receptor sequencing; TCR, T cell receptor; TCR-seq, TCR sequencing.

T cell states and phenotypes

Pre-manufacture T cell features.

The heterogeneity of source T cells has a crucial role in determining the subsequent in vivo persistence of the autologous CAR product^28,30,192 (Fig. 3a). For example, an integrative bulk and single-cell profiling study of pre-manufacture T cell populations revealed that a higher proportion of naive and early memory T cells, especially in the CD8⁺ compartment, is associated with longer CAR T cell persistence and greater efficacy in patients with B cell malignancies²⁸. Moreover, TCF7 (TCF1) network activity in effector T cells was found to correlate with long-term CAR T cell persistence, whereas upregulation of genes associated with the type I interferon response, such as RSAD2, IRF7, MX1, ISG15, OASL and IFIT3, was associated with poor CAR T cell persistence.

CAR T cell features.

The composition of the CAR T cell products infused into patients is also highly heterogeneous, and specific CAR T cell populations have been associated with favourable or unfavourable responses^{15,30,97,117,193,194} (Fig. 3a). For example, bulk transcriptomics profiling of CAR T cells before infusion into 41 patients with advanced-stage chronic lymphocytic leukaemia (CLL) revealed that patients with a subsequent complete response (CR) had CAR T cells enriched for early memory programmes and expression of IL-6–STAT3 pathway genes³⁰. By contrast, CAR T cells of patients who had only a partial response or of non-responders were enriched for the expression of genes associated with T cell differentiation and exhaustion, aerobic glycolysis and apoptosis³⁰. Features of the CD8⁺ CAR T cell subset are crucial determinants of therapeutic efficacy, especially at the initial stage of the antitumour response. In a single-cell multidimensional analysis of CAR T cell infusion products received by patients with relapsed and/or refractory LBCL, enrichment for memory CD8⁺ CAR T cells was associated with a CR at 3 months, whereas CAR T cell products enriched for exhausted CD8⁺ T cells and with low levels of TCR clonotypic diversity were associated with a partial response or progressive disease¹⁵.

The composition of CAR T cells also affects therapeutic persistence and/or long-term disease remission. In a study involving 12 patients with B-ALL, combined scRNA-seq and cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq) of the CD19-directed CAR T cell infusion products indicated that deficiency of T helper 2 cell function and enrichment for effector-related phenotypes is predictive of CD19⁺ disease relapse¹⁹⁴. Furthermore, data suggest that certain CD4⁺ CAR T cell states are a predominant requirement for long-term remission following CD19-directed CAR T cell therapy. In a longitudinal study, both bulk and single-cell multidimensional omics data were used to characterize the functional and molecular features of CD19-directed CAR T cells in two patients with CLL who had durable CRs lasting more than a decade; persistent CD4⁺ CAR T cell clones with cytotoxic, proliferating and metabolically active phenotypes were identified as a key determinant of long-term CAR T persistence⁹⁷.

CAR T cell-intrinsic features, including expression of specific genes^{29,32,195,196} (such as key transcription factors), epigenetic signatures³³ and metabolic states³⁴, are associated with therapeutic efficacy (Fig. 3a). For example, single-cell transcriptomics and chromatin accessibility profiling data implicate the transcription factors NR4A1, NR4A2 and NR4A3 in CAR T cell dysfunction²⁹. Moreover, ATAC-seq of Nr4a triple-knockout CAR T cells showed enrichment of accessible chromatin regions containing binding motifs for the transcription factors NF-κB and AP1 (ref. ²⁹). Data from a multidimensional transcriptomics, epigenomics and single-cell proteomics study indicate that the transcription factor BATF can prevent CAR T cell exhaustion and facilitates the transition of phenotypes and transcriptional profiles towards an effector-like state¹⁹⁶. In an alternative approach to gene expression analysis, a predictive DNA methylation signature termed EPICART has been generated using DNA methylation array data from 114 patients with B cell malignancies treated with CD19-directed CAR T cells³³. Pre-infusion CAR T cells positive for this signature were associated with better clinical outcomes in both the initial discovery cohort (n = 77) and an independent validation cohort (n = 35) and were enriched for naive-like or early memory T cell phenotypes³³.

The metabolic state of the CAR T cells might influence antitumour activity and toxicity (Fig. 3a). A guide RNA-based CRISPR gain-of-function screening platform has been developed and applied to primary CD8⁺ T cells, resulting in the identification of upregulation of PRODH2, which encodes hydroxyproline dehydrogenase (an enzyme involved in proline metabolism), as a means to augment CAR T cell function³⁴. Transcriptomics and metabolomics profiling of PRODH2-knock-in CAR T cells revealed enhanced metabolic and immunological activity, and these cells had improved antitumour activity in multiple mouse models³⁴. Furthermore, CAR T cell metabolic activity might lead to imbalances, such as hypophosphataemia, that contribute to ICANS¹⁹⁷.

Integration of the CAR transgene into the genome can affect key genes, thereby influencing the clinical outcome of CAR T cell therapy^62,198 (Fig. 3a). For example, insertion of a CAR sequence into T cells from a patient with CLL disrupted the transcription of TET2 (encoding a methylcytosine dioxygenase involved in epigenetic regulation) and resulted in an enhanced antitumour response⁶². Further transcriptomics and epigenomics analyses revealed that the patient had a pre-existing hypomorphic mutation in the second TET2 allele and that TET2-disrupted CAR T cells had increased expression of effector molecules, such as perforin and granzyme B, as well as a central memory state that prevented cells from becoming terminally exhausted. In a high-throughput sequencing study of CAR vector-integration sites in 40 patients with CLL or B-ALL, integration sites in patients who responded to therapy were enriched in genes related to pathways mediating T cell proliferation, such as phosphatidylinositol, cyclic AMP or TCR signalling, or covalent chromatin modification, suggesting that insertional mutagenesis in these genes might lead to better clinical outcomes¹⁹⁸.

The intracellular signalling domain of the CAR construct can also influence CAR T cell activity (Fig. 3a). In a simultaneous examination of the dynamics of immunoreceptor tyrosine-based activation motif (ITAM) phosphorylation within the CD3ε, CD3δ, CD3γ and CD3ζ signalling subunits of the TCR using quantitative mass spectrometry, high levels of mono-phosphorylation and a basic residue-rich sequence of CD3ε were found to restrain TCR signalling¹⁰⁹. Notably, addition of the CD3ε-derived ITAM–basic residue-rich sequence region into a CD19–CD28–CD3ζ CAR construct resulted in a CAR T cell product associated with reduced levels of cytokine release, enhanced antitumour responses and increased persistence in a mouse xenograft model of B cell lymphoma. Indeed, different CAR T products have different phenotypes and expansion characteristics. Axicabtagene ciloleucel (axi-cel) and tisagenlecleucel, for example, differ not only in the source of their intracellular co-stimulatory domains (CD28 versus 4–1BB) but also in their hinge–transmembrane domains (CD28 versus CD8α) and their integration vectors (gammaretrovirus versus lentivirus)¹⁹⁹. In a scRNA-seq analysis of the infusion products and 105 peripheral blood mononuclear cell samples collected at different pre-treatment and post-treatment time points from 19 patients receiving axi-cel and 13 receiving tisagenlecleucel, expansion of central memory CD8⁺ cells was only associated with responses to tisagenlecleucel; responders to axi-cel had more heterogeneous T cell populations, while enrichment of CAR-expressing regulatory T (T_reg) cells in the axi-cel infusion product were associated with a lack of response²⁴. Together, these findings underscore the application of multi-omics data in characterizing the differences between distinct CAR designs and cell products.

Tumour cell characteristics

Antigen-dependent resistance.

Tumour cell characteristics, including both antigen-dependent and antigen-independent mechanisms, are crucial drivers of resistance to CAR T cells^166,200,201 (Fig. 3b). Antigen escape in tumours (that is, loss of the target epitope) is the leading cause of antigen-dependent resistance to CAR T cells and has been shown to occur via downregulation²⁰², mutation¹⁶⁰, alternative splicing^81,203,204, lineage switching^205,206 or antigen masking²⁰⁷. A DNA-seq and RNA-seq analysis found that all 12 evaluated patients relapsing with CD19-negative B-ALL after CD19-directed CAR T cell therapy had somatic mutations in CD19 exons 2–5, which are postulated to result in truncation of CD19 and subsequent loss of expression on the cell surface¹⁶⁰. An RNA-seq study demonstrated alternative splicing of CD19 with exon 2 skipping, leading to a loss of the extracellular CD19 epitope⁸¹. By contrast, scRNA-seq in a patient with relapsing B-ALL revealed antigen loss of CD19 via mis-splicing that resulted in a non-functional CD19 transcript retaining intron 2, and this subclone was present before CAR T cell treatment²⁰³. A study using flow cytometry reported switching of B-ALL to a CD19^– myeloid lineage in four patients after CD19-directed CAR T cell therapy²⁰⁵. Combined RNA-seq and ChIP-seq data from mouse models provided further validation that the myeloid lineage switch was triggered following CD19-targeted CAR T cell infusion²⁰⁵. In addition, combined genomics and transcriptomics sequencing has uncovered a novel but rare mechanism of epitope masking related to resistance to CAR T cell therapy: the CAR gene was transduced into a single leukaemic B cell by accident, and expression of the CD19-directed CAR on the progeny of this leukaemic clone prevented recognition by, and thus conferred resistance to, CAR T cells via competitive binding to CD19 on leukaemic cells in cis²⁰⁷. These results demonstrate the complexity of antigen escape mechanisms in B-ALL. However, despite approximately 28% of LBCLs having reduced CD19 expression at relapse after CD19-directed CAR T cell therapy²⁰², mutation and alternative splicing are not observed and the mechanisms remain to be defined; thus, different mechanisms of antigen escape are likely to exist in distinct malignancies. The application of genomics, transcriptomics and/or proteomics profiling will be important to elucidate the mechanisms of antigen escape across diverse contexts and to develop strategies to circumvent antigen-dependent resistance, for example, through the design of dual-targeted CAR T cell products^181,182.

Antigen-independent resistance.

Antigen-independent resistance resulting from the intrinsic insensitivity of tumour cells to the mechanisms of CAR T cell-mediated cytotoxicity also contributes substantially to treatment failure (Fig. 3b). Emerging evidence indicates that dysfunctional death receptor signalling in tumour cells can lead to resistance to CAR T cell cytotoxicity. In a genome-wide CRISPR–Cas9 knockout screen, disruption of genes involved in pro-apoptotic death receptor signalling (including FADD, BID, CASP8 and TNFRSF10B) in B-ALL cell lines conferred resistance to CD19-directed CAR T cells, which was further reinforced by induction of CAR T cell dysfunction as a result of antigen persistence⁵⁵. Conversely, knocking out anti-apoptotic genes (such as CFLAR, TRAF2 and BIRC2) in leukaemic cells increased their susceptibility to CAR T cell-mediated cytotoxicity⁵⁵. Similarly, another CRISPR–Cas9 screen demonstrated that expression of the death receptor Fas on tumour cells is essential for antigen-dependent CAR T cell-mediated killing as well as for bystander T cell-mediated killing of antigen-negative tumour cells²⁰⁸. Notably, DNA-seq and RNA-seq data indicate that patients with LBCL harbouring FAS deletions⁴⁵ or with low FAS expression²⁰⁸ have inferior survival outcomes following CD19-directed CAR T cell therapy.

Other genomics alterations within tumour cells have also been correlated with poor outcomes following CAR T therapy. Integrated targeted DNA-seq of pre-treatment tumour samples from 82 patients with relapsed and/or refractory LBCL receiving CD19-directed CAR T cell therapy demonstrated that TP53 aberrations (mutations and/or copy-number losses) are associated with reduced overall survival but not with progression-free survival⁴⁴. Bulk transcriptomics data from a separate cohort of 562 patients further suggested that TP53 alterations in tumours impair processes such as interferon signalling, death receptor signalling and CD8⁺ T cell infiltration, which are important for CAR T cell-mediated cytotoxicity⁴⁴. TP53 alterations are also linked with increased genomics complexity in a variety of contexts, including LBCL²⁰⁹, and high levels of genomics complexity have been associated with poor outcomes in patients with LBCL receiving CD19-directed CAR T cells^45,46. Taken together, these studies emphasize that integration of multidimensional omics data is crucial to reveal the tumour-intrinsic mechanisms of resistance to CAR T cell therapy.

The TME

The TME typically contains diverse immunosuppressive cells, including T_reg cells, cancer-associated fibroblasts, tumour-associated macrophages and myeloid-derived suppressor cells, and cytokines (such as TGFβ, IL-10, IL-4 and VEGF) that might limit CAR T cell infiltration, proliferation and effector function, and ultimately lead to CAR T cell exhaustion^16,210–212. The challenges to CAR T cell therapy posed by the immunosuppressive TME and the engineering strategies aiming to overcome them have been extensively reviewed elsewhere^166,213. Interrogation of the TME in patients treated with CAR T cells has been limited to targeted gene expression assays^25,210 and has not yet been comprehensively explored with omics approaches, although the findings suggest that intratumoural myeloid cells and cytokine levels are associated with response to CD19-directed CAR T cells. Here, we focus on the application of multidimensional omics data to obtain an in-depth understanding of the hostile TME in order to facilitate the development of new approaches to enhance the intratumoural abundance, persistence and activity of CAR T cells and thereby improve their clinical efficacy (Fig. 3c).

Increasing CAR T cell infiltration.

Omics data have been utilized to identify key regulators that could be leveraged to increase CAR T cell infiltration into the TME. In a mouse model of breast cancer, the stimulator of interferon genes (STING) agonist DMXAA greatly enhanced CAR T cell recruitment to and persistence in the TME, with scRNA-seq revealing favourable shifts in the chemokine milieu and the balance of immunostimulatory versus immunosuppressive myeloid cells²¹⁴. In a kinome-wide screening study, RNA-seq revealed that PAK4 has an important role in regulating mesenchymal-like transcription activation in glioblastoma-derived endothelial cells; inhibition of this kinase improved T cell infiltration and sensitized glioblastoma tumours to EGFRvIII-directed CAR T cells in a mouse model³⁵.

Strengthening CAR T cells against immunosuppression.

Multi-omics analyses also present a powerful approach for identifying key immunosuppressive factors in the TME, thereby enabling the engineering of CAR T cells capable of resisting such factors¹. A single-cell transcriptomics analysis of oesophageal squamous cell carcinomas indicated that PD-L1 is upregulated on tolerogenic dendritic cells in the TME²¹⁵, and data from a pan-cancer scRNA-seq atlas of T cells suggested that binding of PD-L1 to PD-1 (ref. ²¹⁶) inhibits T cell cytotoxicity and facilitates immune evasion, underscoring the known role of the PD-L1–PD-1 axis as a key mediator of immunosuppression in the TME. Therefore, the efficacy of CAR T cells against poorly responding tumours might be enhanced by combining CAR T cells with anti-PD-(L)1 antibodies²¹⁷, engineering PD-1-blocking CAR T cells²¹⁸ or CAR T cells that express PD-1 dominant-negative receptors²¹⁹, or genetically disrupting PD-1 expression in CAR T cells²²⁰.

A pan-cancer analysis based on RNA-seq data from TCGA and GTEx revealed that FASLG (encoding Fas ligand; FasL) is often overexpressed in the TME²²¹. Accordingly, CAR T cells co-engineered to express dominant-negative variants of Fas (the receptor of FasL that is typically highly expressed on autologous T cells used for adoptive cell therapy) had increased persistence and antitumour efficacy in mouse models of various solid and haematological malignancies²²¹. In addition, a CyTOF analysis demonstrated that IL-18-secreting CAR T cells can modulate the TME and enhance the activity of endogenous immune cells (for example, promoting endogenous CD8⁺ T cells with a central memory phenotype, macrophages with an M1 phenotype, and dendritic cells with a more mature and activated phenotype) in diverse mouse models, thereby promoting antitumour immune responses²²². Taken together, these studies demonstrate that leveraging multidimensional omics data can greatly help to overcome factors in the TME that limit the effectiveness of CAR T cell therapy for solid tumours.

The microbiota

The microbiome is emerging as a crucial factor affecting the efficacy of CAR T cell therapy (Fig. 3d). In a multicentre study involving a prospective cohort of 48 patients with B cell lymphomas or leukaemias receiving CD19-directed CAR T cell therapy, 16S rRNA-seq and metagenomics shotgun sequencing of the baseline faecal samples revealed a considerably lower level of α-diversity and a substantially different composition of the gut microbiota compared with that of individuals without cancer¹⁴¹. Furthermore, specific bacterial taxa of Lachnospiraceae, Ruminococcaceae and Bacteroidaceae were in greater abundance in patients who had a CR at day 100 than those who did not. Preclinical data from mouse models of melanoma and pancreatic cancer indicate that two microbial metabolites identified using metabolomics data from the human commensal bacteria Megasphaera massiliensis, namely pentanoate and butyrate, can increase the antitumour activity of CAR T cells through metabolic and epigenetic reprogramming²²³. Prior knowledge that not only the intestinal microbiota but also the intratumour microbiota can have multiple effects on human cancers along with the development of technologies that enable quantitative analyses of the microbiome^132,142 will facilitate further investigations to characterize the influence of the microbiota on the efficacy of CAR T cell therapy.

Minimizing CAR T cell-related toxicities

The toxicities of CAR T cell therapies, which can be serious or even fatal¹⁸, remain a substantial challenge to widespread clinical application^224–226. In the process of killing antigen-expressing target cells, CAR T cells induce a series of inflammatory responses via the secretion of various cytokines and chemokines — a double-edged sword that amplifies antitumour immune responses but can also cause severe toxicities¹⁶. The most common CAR T cell-induced adverse events include CRS, ICANS and prolonged cytopenias²²⁷. Several studies have demonstrated the potential of multi-omics profiling to elucidate the mechanisms underlying these toxicities and thus inform potential prophylactic and therapeutic strategies.

CRS

CRS results from robust cytokine production by CAR T cells following engagement with target cells as well as by subsequently activated endogenous immune cells, which creates a loop of immune activation that can cause a ‘cytokine storm’^226,228. Characterization of cytokine levels and immune cell interactions in the context of CRS has been crucial to understanding its aetiology (Fig. 4a). Transcriptomics approaches have been used to determine the key immune cell types and/or cellular interactions that mediate this excessive cytokine release. In a mouse model of CAR T cell-induced CRS, bulk RNA-seq of myeloid cells isolated by fluorescence-activated cell sorting revealed that increased expression of IL-1 receptors and IL-6 by these cells, and particularly by macrophages, is associated with greater severity of CRS³⁷. Moreover, a scRNA-seq analysis of CD45⁺ leukocytes from a humanized mouse model of high-burden leukaemia implicated circulating human monocytes as the primary source of IL-6 and IL-1 released following CAR T cell infusion, with a time-course analysis indicating that production of IL-1 preceded IL-6 during the development of CRS³⁶. Furthermore, a cytokine signature generated through proteomics profiling of serum samples from 51 patients receiving CD19-directed CAR T cell therapy accurately predicted the likelihood of severe CRS, potentially enabling early intervention²²⁹.

Fig. 4 |. Summary of factors related to major categories of CAR T cell-induced toxicities revealed by multi-omics data.

a, Factors involved in chimeric antigen receptor (CAR) T cell-associated cytokine-release syndrome (CRS) that have been characterized through transcriptomics and proteomics. b, Factors related to the risk and/or mechanism of immune effector cell-associated neurotoxicity syndrome (ICANS) have been identified using transcriptomics and microbiomics data. c, On-target, off-tumour effects of CAR T cell products can also be characterized, and potentially predicted, through transcriptomics and proteomics analyses. T_reg, regulatory T.

ICANS

ICANS is another major toxicity of CD19-directed CAR T cell therapy^19,227,230. Targeted gene expression analysis has identified an association between a high pre-treatment intratumoural density of T_reg cells and reduced risk of ICANS in patients with LBCL²⁵, and targeted single-cell cytokine analysis has shown that higher frequencies of polyfunctional IL-17A-producing T cells within infusion products are associated with increased risk of high-grade ICANS. Moreover, scRNA-seq of CD19-directed CAR T cell infusion products for patients with LBCL revealed a small subset of cells with a monocyte-like transcriptional signature within the infusion products of patients with high-grade ICANS¹⁵. Notably, these cells express high levels of IL-1β; this finding, together with the independent aforementioned support for the pathophysiological role of IL-1 signalling in CRS^36,37 and ICANS^15,36, has prompted clinical investigation of an IL-1 receptor antagonist for prophylaxis of CRS and ICANS (NCT04432506). In addition, scRNA-seq data from human brain tissue suggest that CD19 expression on brain mural cells might also be a target for CD19-directed CAR T cells⁸⁹. Furthermore, 16S rRNA-seq and metagenomics shotgun sequencing of stool samples from 45 patients with B cell malignancies receiving CD19-directed CAR T cells demonstrated an association between ICANS and faecal microbiota composition¹⁴¹. The mechanisms underlying ICANS are multifactorial and remain poorly understood; however, a clearer understanding might be obtained by leveraging multi-omics data (Fig. 4b).

On-target, off-tumour toxicities

The targets of most CAR T cell products currently used in the clinic and/or being tested in clinical trials are expressed to some extent on non-malignant cells; therefore, CAR T cells can cause damage to non-malignant tissues and result in on-target, off-tumour toxicities^27,38,227, especially when targeting solid tumours¹⁸. Predicting the risk of on-target, off-tumour toxicities and monitoring patient safety through multi-omics profiling during treatment could potentially enable more efficient management of such adverse effects³⁸. By combining bulk transcriptomics and proteomics data, a targetable landscape encompassing >100 single CAR targets and >100,000 logic-gated target pairs with minimal predicted toxicity has been reported³⁸. Similarly, the study of two independent large-scale scRNA-seq datasets⁹⁰ defined the safety landscape of 591 CAR targets and 320,884 logic-gated “AND” design target pairs at the single-cell level. Notably, by leveraging scRNA-seq to dissect rare cell populations and/or types, several potentially risky targets were identified that might result in toxicities affecting several important organs, including EGFR, prostate stem cell antigen and VEGFR2, which were not identified as a safety risk in bulk expression analyses⁹⁰. In summary, multi-omics analyses could substantially broaden molecular insights on toxicities and thus facilitate the development of approaches to mitigate the adverse effects of CAR T cell therapies (Fig. 4c).

Conclusions

CAR T cell therapy has transformed the management of B cell malignancies, yet substantial numbers of patients do not have a clinical response or have disease relapse owing to poor CAR T cell function and/or resistance of tumour cells to such therapies. Moreover, the applicability of CAR T cell therapies remains limited owing to a series of challenges that include a lack of appropriate CAR targets, the potential for disease relapse and treatment-associated adverse events. Solutions to these challenges are under exploration and development. The expanding multi-omics data repertoire could be utilized to obtain new molecular insights into CAR T cell therapy. Multi-omics profiling of tumour and non-malignant tissues provides a feasible strategy for large-scale screening of CAR targets and logical designs. Meanwhile, characterization of T cells, tumour cells, the TME and the microbiota can decipher the molecular features and pathways that determine therapeutic efficacy as well as the potential underlying mechanisms and pivotal regulators of different categories of toxicities associated with CAR T cell therapy. The current FDA-approved CAR T cell products are manufactured from autologous T cells owing to the advantage of avoiding graft versus host disease, yet such products have disadvantages related to long manufacturing times and impaired functional fitness of patient-derived T cells²³¹. Notably, multi-omics studies of donor-derived allogeneic CAR T cell products and their comparison with autologous products using transcriptomics, proteomics and metabolomics data have demonstrated the impaired metabolic fitness of autologous CAR T cells as well as a higher level of activation in allogeneic CAR T cells, indicating a greater antitumour potential^87,130,232. These studies have also advanced the mechanistic understanding of allogeneic products and delineated a promising future for ‘off the shelf’ CAR T cells.

Advances in single-cell technologies have greatly facilitated exploration of the remarkable diversity of cellular phenotypes and states among both tumour and immune cells in the TME, expanding our collective knowledge in the field of immuno-oncology^93,233. The parallel improvements in analytical algorithms to interpret single-cell multi-omics data have supported important advances in single-cell technologies^{70,113,234–237}. Further application of single-cell technologies to study various aspects of CAR T cell therapy is likely to provide rich and high-resolution information, offering clinically relevant insights into treatment strategies to maximize the potential for durable remissions. Furthermore, spatial profiling technologies enable in situ cell characterization and mapping of spatially preserved tissue sections. Although not yet utilized to study CAR T cell therapy, spatial transcriptomics and proteomics have been used to investigate interactions between tumour cells and infiltrating immune cells in the context of immune-checkpoint inhibitor therapy in order to explore the clinical relevance of cell localization^238,239. These advances in single-cell and spatial multi-omics technologies have demonstrated promise in dissecting the cellular components and their intercellular connections, providing opportunities to overcome obstacles to effective therapeutic targeting, including immune evasion and immunosuppression of CAR T cells.

Leveraging valuable multidimensional resources through multi-omics data integration is also of great relevance in elucidating the evolution and transitions of cells during therapy. For example, combined RNA-seq, CITE-seq and single-cell ATAC-seq data have been used to track the molecular determinants of long-term CAR T cell persistence²⁸. Furthermore, the integration of omics data from multiple studies enables a substantially larger sample size and might thus lead to novel discoveries of important biological processes or features formerly underappreciated in studies with relatively small sample sizes²⁴⁰. To this end, further optimization of computational frameworks for omics data integration is required to fully leverage multi-omics data in future research in the realm of CAR T cell therapy and beyond^164,241.

Key points.

• Multidimensional omics data, encompassing genomics, epigenomics, transcriptomics, T cell receptor-repertoire profiling, proteomics, metabolomics and microbiomics, have been exploited to advance our mechanistic understanding of chimeric antigen receptor (CAR) T cell therapy.

• Utilization of multi-omics data derived from both tumour and non-malignant tissues at the bulk and/or single-cell levels is a powerful approach to identifying optimal targets for highly efficacious and safe CAR T cell therapy.

• Integration of bulk and/or single-cell multidimensional omics data has been applied to investigate key determinants of CAR T cell persistence and antitumour efficacy, including T cell states and phenotypes, tumour cell characteristics, the tumour microenvironment and the microbiome.

• Leveraging multi-omics data promises to elucidate the mechanisms underlying CAR T cell-related toxicities, including cytokine-release syndrome, immune effector cell-associated neurotoxicity syndrome and on-target, off-tumour toxicities.