Applications of single‐cell omics for chimeric antigen receptor T cell therapy

Chimeric antigen receptor (CAR) T cell therapy is a promising cancer treatment modality. The breakthroughs in CAR T cell therapy were, in part, possible with the help of cell analysis methods, such as single‐cell analysis. Bulk analyses have provided invaluable information regarding the complex molecular dynamics of CAR T cells, but their results are an average of thousands of signals in CAR T or tumour cells. Since cancer is a heterogeneous disease where each minute detail of a subclone could change the outcome of the treatment, single‐cell analysis could prove to be a powerful instrument in deciphering the secrets of tumour microenvironment for cancer immunotherapy. With the recent studies in all aspects of adoptive cell therapy making use of single‐cell analysis, a comprehensive review of the recent preclinical and clinical findings in CAR T cell therapy was needed. Here, we categorized and summarized the key points of the studies in which single‐cell analysis provided insights into the genomics, epigenomics, transcriptomics and proteomics as well as their respective multi‐omics of CAR T cell therapy.


INTRODUCTION
Adoptive cell therapy (ACT) is a branch of cancer immunotherapy that harnesses the strength of immune cells, mainly T cells, to eradicate tumour cells.Chimeric antigen receptor (CAR) T cells are an offshoot of ACT, which use CAR constructs to specifically target a designated antigen.CAR T cell therapy reinvigorated the field of cancer immunotherapy with the FDA approval of its products for multiple haematologic malignancies, such as Yescarta ® and Kymria ® in 2017 [1].CARs constitute an extracellular antibody single-chain variable fragment Abbreviations: ACT, adoptive cell therapy; ATAC-seq, assay for transposase-accessible chromatin using sequencing; CAR, chimeric antigen receptor; CITE-Seq, cellular indexing of transcriptomes and epitopes by sequencing; CR, complete response; CRS, cytokine release syndrome; CyTOF, cytometry by time-of-flight; FACS, fluorescenceactivated cell sorting; GvHD, graft-versus-host disease; ICANS, immune effector cell-associated neurotoxicity syndrome; ICB, immune checkpoint blocker; MDSC, myeloid-derived suppressor cells; MHC, major histocompatibility complex; MRD, minimal residual disease; scDNA-seq, single-cell DNA sequencing; scRNA-seq, single-cell RNA sequencing; TAA, tumour-associated antigen; TCR, T cell receptor; TF, transcription factor; TIL, tumour-infiltrating lymphocyte; TME, tumour microenvironment; WES, whole exome sequencing.
(scFv) domain, a transmembrane domain, one or more intracellular, co-stimulatory domains (e.g., CD28 or 4-1BB) depending on the CAR generation, and an intracellular signalling domain (CD3ζ) to activate the CAR T cell upon recognizing the target antigen and exert its cytotoxic function [2].Peripheral blood mononuclear cells (PBMCs) are isolated from blood and then genetically modified to express the CAR.The transfer of CAR-coding gene into host T cell genome is done using γ-retroviral or lentiviral vectors, electroporation or using CRISPR/Cas9 or other gene-editing platforms for a precise integration [3].The CAR T cells are then expanded and infused into the patient where, upon antigen binding, activate and differentiate into effector T cells, some of which go on to become memory T cells for long-term immunity [4][5][6].Despite the high cost of CAR T therapy, efforts are being made to turn them into a mainstream modality of cancer treatment.There is, however, still room for improvement.
The outcome of ACT depends on cell product interactions with the tumour microenvironment (TME) and their molecular traits.Only by gaining knowledge about these traits can we devise strategies to circumvent TME immunosuppressive environments or boost cell products.Single-cell analysis is an emerging tool that can shed light on the complexities of a heterogeneous tumour on a single-cell level.Tumours are highly heterogeneous niches that encompass genetic variations in the cells themselves [7].Each cluster of tumour cells is unique in its molecular features.These intra-tumoural variations arise due to several intrinsic and extrinsic factors, including stochastic genetic mutations, cell differentiation and the diverse cellular population of TME, resulting in a complex disease with multiple layers of heterogeneity [8].The resultant clones (also subclones), which are shown to have 10 000 mutations on average, greatly determine the outcome of the treatment [9].High degrees of clonal diversity can affect treatment outcomes.An undetected subclone may have been resistant to treatment in a patient, causing a relapse as a result.It is therefore possible for targeted therapies, which treat tumours a homogeneous tissue, to ultimately fail to eradicate the disease.A single-cell analysis of each tumour's heterogeneity is consequently essential.Single-cell analysis includes a wide array of different features that illuminate the traits of any small cell populations in a large cell pool [10].The subsequent understanding of the role of each tumour cell population in tumorigenesis, metastasis and drug resistance is equally important.Here, we reviewed the experiments that conducted in-depth, single-cell analysis on CAR T cells and the TME in order to provide information that will assist in boosting CAR T cell therapy.

SINGLE-CELL ANALYSIS REFINES CAR T CELL THERAPY
The conventional cancer treatment modalities are chemotherapy, radiotherapy and surgery alone or combined.With cancer being the second leading cause of death worldwide, preclinical and clinical studies are searching for more effective, advanced treatments, including immunotherapy [11].As a branch of cancer immunotherapy, ACT is showing promising results in the clinic.Although ACT has become popular with its recent strides, characterizing target tissue, recognizing new targets in heterogeneous disorders, and evaluating the efficacy of treatments refine CAR T cell therapy.Early trials of CD19-directed CAR T cells demonstrated complete response (CR) in 60% [12] to 83% [13] of relapsed or refractory (r/r) B cell acute lymphoblastic leukaemia (B-ALL), 28% [14] to 71% [15] of r/r chronic lymphocytic leukaemia (CLL), and almost half of the non-Hodgkin's lymphoma (NHL) patients [16,17].Still, multiple tumour-intrinsic and -extrinsic factors hinder this approach.CAR T cell efficacy is contingent upon recognizing a suitable tumour-associated antigen (TAA), an antigen that is ideally expressed on tumour cells and scarcely found on healthy cells.The resultant 'on-target on-tumour' toxicity eliminates tumour cells.As an unfortunate side effect, the on-target off-tumour effect could cause organ damage as well [18].Immune effector cellassociated neurotoxicity syndrome (ICANS) is a serious condition that possibly occurs due to CAR T cell infiltration in the brain.If T cell response is overly aggressive, the subsequent massive release of inflammatory T cell cytokines upon target recognition can promote a systemic cytokine release syndrome (CRS) [19].Limited expansion and short-term CAR T cell persistence also brings about an unsuccessful treatment.The long-term presence of memory T cells prevents the minimal residual disease (MRD) from causing relapse, a life-threatening issue that often occurs 3-6 months after CR in more than half of the patients [20].The tumour itself further complicates the treatment.Tumour mutations at the beginning or later stages of the treatment, in which loss of CAR target antigen could occur, is a cause for primary refractoriness or relapse of the disease, respectively [21].The heterogeneous nature of tumours only exacerbates this issue.In a heterogeneous tumour of multiple clones that initially responds to treatment, the dominant clones dwindle upon treatment only to be replaced by pre-existing minority clones, decreasing the remission period.The now-major relapsed clones might even be refractory to the initial treatment by not expressing the CAR-directed antigen [22].The TME is also a cause for concern with plenty of immunosuppressive features at its disposal.This antagonistic milieu can boost CAR T cell exhaustion and limit their persistence, house myeloid-derived suppressor cells (MDSC) and regulatory T cells (Tregs), repress CAR T cells with abundant immune checkpoint molecules and alter their omics profile [23][24][25][26].
Before single-cell analysis, bulk sequencings were the standard for omics analyses.Both technologies have unique advantages and can be used in tandem.The main advantage of single-cell sequencing is that it provides information on individual cells rather than measuring expressions across a cell population.This allows it to reveal cell heterogeneity and subclones in a heterogeneous population of cells.The analysis of immune or tumour cells on a single-cell scale can provide information about the inner workings of these cells, their crosstalks and how they shape the treatment outcome.Single-cell omics analyses have a generally similar workflow: single-cell isolation, extraction and amplification of the desired biological molecule or sequence, construction of sequencing libraries and sequencing, and finally bioinformatic data analysis.A joint analysis of two or more of cell biological molecules is called multi-omics [27].Single-cell analysis can be geared towards CAR T cell products, TME or any other tissue involved in the treatment process, and the data can be used in different scenarios, such as finding a correlation with treatment outcome, designing atlases of tumour or CAR T cells or their molecular alterations, and so forth.Some applications of single-cell analysis in the context of CAR T cell therapy are: establishing the mechanisms of CAR T cell therapy resistance in either immune cells or TME; revealing the effect of CAR gene integration on T cells efficacy; uncovering the mechanisms of CAR T cell toxicity; exposing and comparing TME characteristics before and after CAR T therapy; drawing correlations between the omics profile of CAR T or tumour cells and drug resistance, relapse or clinical outcome; discovering CAR T-or TMErelated factors that cause CAR T cell exhaustion; evaluating the quality of CAR T products in vitro; initial screening and selecting patients who would respond better to CAR T therapy; discovering novel cell populations in TME or CAR T cell products; and estimating the risk of CAR T toxicity in different organs (Figure 1) [28][29][30].In the following sections, we will dissect each of the various studies that have applied single-cell analysis genes to proteins to provide an answer to the above-mentioned drawbacks of CAR T cell therapy.

GENOMICS
Single-cell whole-genome/DNA sequencing (scWGS/ scDNA-seq) is a potent tool for assessing clonal evolution and variations in a heterogeneous cell population.Unlike bulk WGS, which homogenizes and pools DNA content of sample cells, scDNA-seq identifies and analyzes cells individually to detect low-level DNA mosaicism [31,32].Despite being an effective platform to discern clonal dynamics or lineage tracing of cancer cells, bulk genome sequencing suffers from multiple pitfalls.The genome, for instance, must be massively multiplied, and the slightest errors hinder this approach.scWGA traditional methods (e.g., MDA, DOP-PCR) have come a long way, however, and the latest ones (e.g., LIANTI, MALBAC) have more genome coverage and are more accurate for single-nucleotide variants (SNVs) and copy number variations (CNVs) [28].CNVs are regions of the genome that are structurally variant between the DNA of cells, so this marker can be used to single out low-population clones of a heterogeneous tumour population that would have been overlooked in bulk analyses.Still, the sequencing portion of genomic analysis is costly because of its large size.This has prompted scientists to sequence and analyse the whole exome or only certain genes with whole exome sequencing (WES) [33].The analysis of the data is also challenging and more prone to errors since allelic dropouts (where an allele is not amplified) and unbalanced amplifications occur, and imputations must be used to fill in the missing data.scDNA-seq adds another layer of complexity because the starting sample is the genetic material of a single cell.This makes genetic nucleic acid degradation, sample mishandling and contamination especially challenging compared to bulk cell analysis.The aim of upcoming research is to circumvent these challenges [34].WGS is applied to tracking mutations in highly heterogeneous tumours following CAR T cell therapy.Balke-Want et al. reported the first investigation of clonal evolution (tumour cells mutations over time) by comparing diffuse large B cell lymphoma (DLBCL) mutations via bulk WES before and after receiving anti-CD19 CAR T therapy and found a correlation between PTPRA mutation and relapse and extranodal lymphoma [35].The information can also help deduce what mutations cause relapse in certain patients.Subclones susceptible to or resistant to treatment can also be distinguished among tumour populations.These results prove how bulk WGS can be used to pinpoint patients with high relapse risk who may require other immunotherapy approaches.Future studies can utilize scDNAseq to investigate the effect of mutations in various tumour cell clones on treatment outcomes.
Although the hurdles of single-cell genome analysis make it less widely used, it is still delivering valuable information to scientists and physicians.This assay has only recently been used in CAR T cell therapy.Genetic alterations are generally rectified in healthy cells, but since this is not the case with cancer cells, analysing and comparing these genetic mutations in them could yield invaluable information regarding cell lineage.As discussed, cancer is a heterogeneous disorder, and CNV involvement as a risk factor in cancer inception, progression and therapy resistance is established [36].The first tumour single-cell sequencing protocol by Navin et al. demonstrated the genomic heterogeneity of the tumour cell populations, providing insight into cancer evolution.After isolating single cells via fluorescence-activated cell sorting (FACS) in a mixed population of breast cancer cells, WGS was implemented [37].Then, the analysis of 100 single cells in a polygenic tumour revealed three distinct subclones that would have been overlooked in a bulk-cell analysis.Developing methods are greatly increasing the throughput of the number of cells scDNAseq can distinguish by using barcoding systems, for example, Evrony et al. [31].
scDNA-seq in CAR T cell therapy is mostly applied to T cell receptors (TCRs) analysis.Because CAR T cells are essentially T cells, they express unique TCRs as well.The alpha and beta chains of a TCR are distinctive in each clone since their coding genes (TRAC, TCRB) are, respectively, a random combination of VJ and VDJ and other random deletions and insertions.The complementary determining region 3 (CDR3) loop of TCRs is unique in that it is created by V(D)J rearrangements.This distinctive structure, which remains constant through differentiation and expansion and gets passed on to daughter T cells, makes TCR chains a molecular marker for tracing T cell clonal dynamics [30,38].TCR sequencing data can be combined with single-cell RNA sequencing (scRNAseq), which divides cells into distinct clusters based on their transcriptional profile, to examine the trajectory of each cell type and find its lineage.For instance, clusters with different transcriptional profiles that have similar TCR α/β sequences are indicative of a common origin.A study uncovered details of the changes in the clonal composition of CD4+/CD8+ CAR T cells (1:1 ratio) by comparing TCRB sequences of the infused CAR T cells with those of Weeks 1 and 4 CAR T cells (NCT01865617) (Table 1) [39].CD8+ CAR T cells became dominant, and their TCRB repertoire and diversity diminished over the month.This means some clones expanded more due to some unknown characteristics.Most of the remaining CD8+ CAR T cells belonged to only 10 clones.Loss of heterogeneity in the infused CAR T cells was not correlated with a particular clinical response.However, the methodology can be used to identify clones with more survivability and toxicity to tumour cells.The combined scRNA-seq and TCR sequencing data illustrated how most of the dominant clones after infusion could be traced back to the pre-infusion clusters with overexpression of genes associated with proliferation and cytotoxicity.The decline in CAR T cell TCRB diversity was also reported in a bulk analysis study as well where they sought to elucidate the effect of CAR T cell therapy on local immunity and host T cells [40].They observed a decline in the TCR repertoire of bulk T cells in peripheral blood and bone marrow after receiving CAR T therapy.These remaining T cells greatly expanded after the CAR T-mediated destruction of leukaemic cells, and their expansion rate was higher than that of the infused CAR T cells [40].These results showed how CAR T cell therapy might have indirectly instigated an immunotherapeutic response in host immune cells as well.In a recent investigation of the traits of effective CAR T cells, one group dissected the gene-expression profile and TCR sequences of roughly 180 000 pre-and post-infusion FACS-sorted CD19 CAR T cells in 14 patients across 6 months [41].Interestingly, only seven CAR T cells remained to be captured in Month 6.As expected, scRNA-seq sequencing showed that pre-infusion CAR T cells expressed higher proliferation (e.g., MKI67), cell cycle (e.g., CDK1) and DNA replication (e.g., MCM7 and TOP2A) genes, and post-infusion cells expressed higher levels of cytotoxic genes (e.g., GNLY, PRF1 and NKG7).Starting at Week 2, the effector subset composed the majority of CAR T cells (65%) and remained the dominant population thereafter.GZMK-expressing cytotoxic CAR T cells consistently made up almost one-quarter of the population throughout the experiment.Two subsets of exhaustion (e.g., TOX, LAG3 and TIGIT) and dysfunctional apoptotic (e.g., CASP8) post-infusion CAR T cells were also identified whose population increased throughout the month.The dysfunctional CAR T cell numbers were elevated during Weeks 3 and 1.This may have resulted from the in vitro to in vivo change in the CAR T cells environment.A surprising finding was TIGIT upregulation in some pre-injected clones.This expression could be useful because they showed that TIGIT+ preinfusion CAR T cells are less exhausted after stimulation.Thus, the use of such clones that provide activated CAR T cells in the later months of treatment could support a durable response.The authors also used TCR α and β sequencing as a natural barcode to track CAR T cell clonotypes across 3 months [41].Clonal size and diversity declined at Week 1 and peaked at Week 3 and kept declining from then on.They then combined these results with the results of scRNA-seq to search for identifiable signatures of effector CAR T cells in the preinfusion product.For example, a cluster of GZMK+ effector cells at an earlier time point shared TCR with a cluster of GZMK-GZMB+ effector cells at a later point, which shows the latter were derived from GZMK+ cells.Because cytotoxic CAR T cell subsets have clear, effectorassociated transcriptional signatures, the authors first T A B L E 1 A summary of genomic and epigenomic studies in CAR T cells.
• TCR sequencing can show clonal diversity and which clones create effector subtypes [41] Autologous CAR T cells CRISPR/Cas9 αHER2-()-CD3ζ scCAR-seq scRNA-seq • SpeedingCARs, a high-throughput system of creating various CARs and testing their cytotoxicity via single-cell sequencing, was made [43] Epigenomics • Exhausted CAR T cell predominantly differentiate from effector CD8+ CAR T cells.
• BATF and IRF4 are pivotal regulating TF in CAR T cell exhaustion with more minor roles for JDP2, BACH1, JUND, and PRDM1 [55] Autologous CAR T cells LV αCD19-4-1BB-CD3ζ EpiVIA (scATAC-seq • EpiVIA can be used to show the correlation between chromatin accessibility and lentiviral integration sites.The results can be used to discern which integrations result in poor or favourable CAR T cell expansion • DNA methylation of some CpGs is associated with CRR, CRS, and neurotoxicity [62] Pre-and post-infusion CAR T cells/ALL LV αCD19-4-1BB-CD3ζ

WGBS
• Loss and gain of methylation promote effector T cell phenotype while suppressing helper T cell trajectory in CAR T cells.
• CAR T cells undergo massive methylation remodelling, and demethylation occurs 10 times more than methylation.
• CAR T cells gradually differentiate into effector and exhaustion phenotypes which is consistent with the DNA methylation of the corresponding genes.
speculated that a limited number of clones in the preinfusion product gave rise to dominant cytotoxic clones.TCR lineage tracking, however, linked the dominant clones to multiple cell clusters in pre-infusion CAR T cells [41].This means that transcriptional profiling of pre-infusion CAR T cells alone is not sufficient to pinpoint the clones that will later become dominant.Hence, by tracing the TCRs of the most effective CAR T cells back to their precursors before infusion, transcriptional signatures that give rise to the most robust CAR T cells can be discovered.The amalgamation of WGS data create algorithms and stratifications that clarify the genetic diversity of cancers to choose the most optimal therapeutic approach.The LymphGen algorithm, for instance, used WGS data from previous studies to classify DLBCL into seven distinct genetic subtypes through a probabilistic model.Each subtype differed in gene expression, intracellular oncogenic pathway, TME and potential therapeutic targets [42].Once extensive single-cell analyses determine the genomic features associated with optimal response in specific CAR T clones, these clones can be handpicked before in vitro expansion from the heterogeneous CAR T population, expanded and infused into the patient.Single-cell sequencing can also help develop more potent CAR constructs.Just recently, the speedingCARs method, integration of signalling domain shuffling plus scRNA-seq and CAR (scCAR-seq) sequencing, was introduced to create a plethora of distinct CAR constructs and identify which ones possess higher cytotoxic functions [43].The method uses a modular cloning strategy to shuffle intracellular signalling domains of CARs to create a library of variant CARs and integrate the CAR into primary T cells' TRAC locus by using CRISPR/Cas9.CAR T cells are then co-cultured with cancer cell lines and FACS-sorted, single-cell analyses characterize the functional properties of each CAR candidate to discern the most optimal CARs for next-generation CAR T cells.CAR integration into the TRAC locus (endogenous TCR knockout) minimizes graft-versus-host disease (GvHD).

EPIGENOMICS
Epigenetics, which regulate gene expressions, is studying the modifications to chromatin without DNA alterations.They include DNA methylation (cytosine methylation and hydroxymethylation), histone modifications and DNA binding to transcription factors (TF) or non-coding RNAs (nc-RNAs) [44].A range of protocols have been developed to analyse each facet of epigenomics, including bisulphite sequencing (for DNA methylation state); scAba-seq (for assessing active DNA methylation); ChIP-seq (for mapping DNA-binding proteins like TFs and histone modifications); DNase-seq and ATAC-seq (for detecting open and accessible DNA regions and mapping TF binding sites); and Hi-C (for showing genomic contacts within DNA to map chromosome folding) [8,45].Like genomic sequencing, epigenomic sequencing is particularly difficult due to the dispersion of epigenomic targets of analysis on only the two copies of genomic DNA.In contrast, transcriptomic sequences exist in dozens of mRNA copies, making analysis easier and studies more abundant.
Generally, chromatin accessibility analytical methods are based on enzyme-based digestion and the isolation of nucleosomes or accessible sites and subsequent NGS.Pioneered in 2013, ATAC-seq (assay for transposaseaccessible chromatin using sequencing) is a sensitive approach to assessing chromatin accessibility, which is also faster than its rival methods [46].It uses a mutant hyperactive Tn5 transposase enzyme to fragment and add sequencing adaptors to open regions of double-strand DNA and tag chromatin active sites.The tagged regions are purified, amplified and sequenced via NGS and analysed to show chromatin open regions and TF binding sites as well.The subsequent modifications, such as adding microfluidics to separate single cells, were implemented for single-cell ATAC-seq (scATAC-seq) [47].Various assays have started using novel indexing methods to analyse hundreds of thousands of cells at single-cell resolutions [48][49][50].scATAC-seq was recently used to investigate the link between chromatin accessibility and CAR T cell exhaustion upon contact with tumour cells.T cell exhaustion is a recognized barrier to adoptive T cell therapy.Physiologically, T cells are meant to be blocked at the end stages of activation to avoid hyperactivation and tissue damage.PD-1 (coded by PDCD1), in concert with CTLA-4, LAG-3, TIM-3, TIGIT, and so forth are receptors that suppress T cell activation and are markers of T cell exhaustion, a phenomenon where T cells become dysfunctional or anergic due to overstimulation [51].In the context of cancer, however, regulatory T cells, macrophages and even tumour cells suppress CAR T cell function and exhaust them by expressing the ligands to these receptors, leading to treatment failure.This makes targeting these receptors with ICIs a viable therapeutic strategy [52].Even before infusion, CAR T cells can become exhausted by the abundance of signalling and growth factors from in vitro expansion [53].ICB cannot always reverse this effect since exhaustion is also determined by DNA epigenetic status [54].In vitro coculture of CD19 CAR T cells and tumour cells can activate T cells, remodel their chromatin accessibility, and differentiate them into distinct clusters [55].Schep and colleagues showed how CD4+ and CD8+ CAR T cells differentiated into memory and effector T cells, respectively.CD8+ CAR T cells also engendered exhausted T cells that had reduced chromatin accessibility in effector genes (IFNG, GZMB) and high chromatin accessibility for co-inhibitory receptor genes (PDCD1, LAG-3) and TFs (TOX2, BATF3).During co-culture, chromatin accessibility changed, and the peak accessibilities were for genes involved in T cell differentiation and activation.Based on chromatin accessibility, the differentiation trajectory of exhausted CAR T cells were linked to CD8+ effector T cells that were differentiated from naïve CD8+ CAR T cells.The authors then measured the relative activity of TFs in each cell type based on their chromatin accessibility using the chromVAR method [56] and combined the results with differentiation trajectories to analyse the remodelling of TFs regulatory networks.The scores of AP-1 and NF-κB transcription factors ubiquitously increased in all T cell differentiation trajectories, but they depreciated when CD8+ effector CAR T cells transitioned to exhausted T cells and were replaced by JUN, NR4A2 and NFKB1 in the intermediate, exhausted CAR T cells (not previously reported by bulk ATAC-seq) and BATF and IRF4 in the terminal, exhausted CAR T cells.In vivo, FACS-sorted scATAC-seq of the chromatin states of B cell maturation antigen (BCMA) CAR T cells of two multiple myeloma patients were also in line with the previous study, and as proof, BATF and IRF4 knockdown reduced PDCD1, CTLA4 and TOX expression, increased the central memory population, and enhanced IFN-γ secretion [55].These findings ascertain what TFs and what cells regulate and differentiate into terminal, exhausted CAR T cells, marking them as a prime target for drug development or gene-editing platforms in order to circumvent T cell exhaustion.
Another facet of CAR T cell therapy where single-cell analysis has proved advantageous is dissecting the effects of CAR integration into genomic DNA.Before CAR T cells are expanded in vitro, CAR transgenes are introduced into their genome.Many clinical productions currently use these random CAR integrations into T cells via viral vectors or electroporation.Since this is an immutable step in CAR T cell production, its effect on the function of the cells must be determined.One case report recounted how CAR transgene integration into the CBL gene greatly expanded the subclones until they constituted almost half of the white blood cell population [57], while another study showed how 94% of CAR T cells 2 months post-infusion originated from one clone whose TET2 was disrupted by CAR transgene [58].Genome profiling to analyse CAR integration has shown that retro and lentiviral transduction results in faulty protein production, which alters CAR T cell function [59].Since poor in vivo expansion and subsequent persistence of CAR T cells in some patients are major therapeutic concerns, one study demonstrated that scATAC-seq Tn5 transposase can be used to add adaptor sequences to viral DNA in the host genome to make locating lentivirus integration sites possible with an accuracy on par with the conventional ligation-mediated polymerase chain reaction (LM-PCR) [60].The customized scATAC-seq assay, named EpiVIA, simultaneously profiled chromatin accessibility and the lentiviral integration site at the single-cell resolution, which could prove useful in CAR T cell settings [60].Considering that integration of CAR transgene into certain genes can affect CAR T expansion and persistence post-infusion, this assay could be used to create an atlas of chromatin state and lentiviral integration sites in CAR T cells to compare these two factors before and after infusion (in multiple time points) in exhausted or persistent clones to unravel, which mechanism underlies poor or favourable in vivo expansion.
DNA methylation is the most prevalent epigenetic modification of the genome in which a methyl group is transferred to the 5 0 carbon of cytosine at CpG sites.DNA methylation controls regulatory DNA elements' accessibility and TF binding, thus governing gene expression.Hypermethylation of tumour-suppressor genes and their ensuing silencing, which uncontrollably expand tumour cells, is a signature in many malignancies [61].CAR T cells possess a distinct gene expression pattern compared to T cells due to epigenetic alterations [44].DNA methylation impacts clinical outcome, CRS, ICANS and overall survival of CAR T cell therapy.One study first profiled the DNA methylation of retrovirally transduced CAR T cells and untransduced T cells before infusion [62].Of the 850 000 inspected CpG sites in the pre-infusion transduced CAR T cells, 984 sites were hypo (465 of 984) or hypermethylated (519 of 984), 339 of which were in regulatory gene regions.After transduction, genes associated with the TCR signalling, cancer development pathways and sister chromatid separation were most affected.This shows how CAR transgene insertion can modify the epigenomic status of CAR T cells.The methylation of 54 of the 984 methylated CpG sites was associated with different clinical outcomes.Finally, the methylation status of 5, 8 and 45 CpG sites correlated with ICANS, CRS and CR, respectively [62].Even though post-infusion CAR T cells were left unanalysed, pre-infusion inspections of these sites can act as a predictive marker to foretell treatment response or potential side effects in advance in other clinical settings.The raw data from the methylation profile of such experiments are also useful for mining in machine learning-based algorithms to discover the correlation between specific gene methylations and CAR T cell functions [63].Besides T cell exhaustion, the presence of long-lived memory T cells that can differentiate into effector cells upon tumour cell reappearance is key for a long-term, relapse-free treatment.Compared to preinfusion CD8+ CAR T cells, post-infusion CD8+ CAR T cells experience massive methylation and demethylation events at CpG sites starting 1 week after infusion, but demethylation is shown to be 10 times more prevalent than methylation events.Most of these events occur at enhancer sites or TF binding sites of genes that govern effector or memory T cell differentiation, indicating the regulatory effect of DNA methylation [64].Over this time period, loss of methylation occurs in cytotoxicity, proliferation and apoptosis pathways (effector T cell traits), which promotes them, while gain of methylation happens in stemness and lymphocyte activation pathways (memory T cell traits), which represses them in the CD8+ CAR T cells.Interestingly, the loss of memory T cell phenotype is associated with progressively increasing methylation signatures of T cell exhaustion genes.The methylation of memory T cell-related genes can also be scored based on their developmental potential [65], and the index can be used to determine the expansion potential of CAR T cells in the weeks following infusion [64].All in all, CAR T cells undergo substantial DNA methylation alterations in vitro and in vivo.These changes regulate various gene expressions and profoundly influence the product's characteristics and treatment outcomes.The role of chromatin accessibility and DNA methylation in CAR T cell exhaustion further illustrates the value of such analyses in reshaping traditional approaches to more precise ones.
Epigenetic evidence of resistance to CAR T therapy can be used to screen patients, assist with the eligibility and selection process for CAR T therapy, and save valuable time before the disease becomes metastatic.Researchers can also utilize these findings to identify novel therapeutic targets for epigenetic drugs to be administered in combination with immunotherapy to produce more robust CAR T cells or reshape TME drug resistance.

TRANSCRIPTOMICS
The emergence of scRNA-seq sequencing marks an exciting era in immunotherapy by offering high-resolution insights into the gene expression and transcriptomic signatures of TME and immune cells that bulk genomics fail to provide [66].scRNA-seq can also extend beyond conventional phenotypic analyses to identify minute expression differences of cells and distinguish subclones into discrete clusters (cell groups with closely similar transcriptional profiles).It can also show the gene expression profile of unstimulated and stimulated CAR T cells and therapy-resistant TME cells.Based on the expression levels of subpopulation marker genes (like naïve, memory, effector, etc.), sample CAR T cells can be classified into various cell clusters.Each cluster contains cells with similar transcriptional profiles.Molecular signatures of the other levels of omics data can help transcriptomics identify even more specific CAR T cell subpopulations.Detailed descriptions of preclinical studies followed by clinical experiments with scRNA-seq technologies in CAR T cells will be the main focus of our discussion here (Table 2).
One of the most basic utilities of scRNA-seq is to stratify cells in specific clusters based on their transcriptional signatures.The population of each CAR T cell subtype, such as memory, effector or exhaustion, changes after antigen stimulation, and scRNA-seq can longitudinally capture these alterations and visualize them in clusters.To explore the heterogeneity of CAR T cell products and the molecular signatures of activated and exhausted CAR T cells, Wang and colleagues used scRNA-seq on more than 50 000 FACS-sorted unstimulated/stimulated PBMC (CAR-non-expressing cells) and unstimulated/ stimulated CAR T cells (CAR-expressing cells) to show how activation alters their transcriptional profile [67].Upon activation, cell distribution in clusters greatly changed, and the cluster with high activation-associated genes (CD25 high CCR7 high SELL low ) massively expanded.These CAR-expressing cells had lower SELL (CD62L) expression and higher granzymes (GZMA, GZMB and GZMH), cytokines (IL3, IL4, IL5, IL8) and activating TFs (FOS) genes.Such transcriptional signatures were found even in CAR-non-expressing cells, likely indicating a role for CAR-expressing cells in boosting the host immune system via cell-to-cell contact.The unstimulated CARexpressing and unstimulated CAR-non-expressing cells had only a few differentially expressed genes in their transcriptomic profiles, and the size of some clusters was insignificantly changed.Although the density of CAR per cell was not clarified, CAR integration was deemed inconsequential to CAR T cell transcriptional profile [67].The exhaustion signature levels varied among donors, suggesting the presence of factors other than stimulation intensity contributing to T cell exhaustion.
Unlike bulk RNA sequencing that offers an average expression profile, scRNA-seq allows the study of heterogeneous cell populations individually.Single-cell analysis is pivotal to deconvolute cell heterogeneity and comprehensively characterize interactions of T cell subtypes in CAR T cell products (CD8+/CD4+, memory and novel subsets).Apart from the genetic and epigenetic alterations previously outlined, transcriptional changes are also implicated in T cell exhaustion.Gene transcription is directly affected by chromatin accessibility.In a multi- • LV αCD19-CD28-CD3ζ 143B osteosarcoma mouse model (infusion products) omics study, the epigenomic/transcriptomic profiles of exhaustion-prone GD2 CAR T cells were compared with non-exhausted CD19 CAR T cells.The main finding was the dysregulation of AP-1 transcription factor in exhausted CAR T cells [68].AP-1 is a family of dimeric TFs in which the conventional c-Jun/c-Fos heterodimers promote IL-2 transcription and T cell expansion.Any other dimerization of a member of this family with c-Jun Abbreviations: ALL: acute lymphoblastic leukaemia; APRIL: a proliferation-inducing ligand, a high-affinity ligand; BA: brexucabtagene autoleucel; BCMA: B cell maturation antigen; BM: bone marrow; CAR: chimeric antigen receptor; CLL: chronic lymphocytic leukaemia; LBCL: large B cell lymphoma; MCL: mantle cell lymphoma; MM: multiple myeloma; NHL: non-Hodgkin lymphoma; PCL, plasma cell leukaemia; TH: T helper.
antagonizes it and favours immunoregulatory expression [69].scRNA-seq/scATAC-seq analyses revealed high expression and enriched TF binding motifs of these antagonizing TFs of the AP-1 family in exhausted CAR T cells.Subsequently, synthetic overexpression of c-Jun not only reduced exhaustion and increased long-term expansion capability, but also enhanced CAR T cells in vivo anti-tumour activity, offering new targets to address exhaustion [68].
A recent study reported on how lentiviral CAR integration affects CAR density and subsequently CAR T cell function [70].The integration profile of CAR High T cells (>5000 CAR/cell) was similar to that of CAR Low T cells (<1500 CAR/cell), and the insertions were mostly at active sites, such as promoters.However, CAR High T cells had almost three times more viral integrations on average.An initial bulk RNA-seq/ATAC-seq analysis by this team showed higher gene expression and chromatin accessibility of genes related to T cell activation and costimulation in CAR High T cells.Subsequent scRNA-seq data demonstrated that in CD4+ and CD8+ populations of CAR High T cells, the signatures of activation genes and tonic signalling were statistically significantly higher than CAR Low T cells.Finally, comparing the scRNA-seq signatures of CAR High T cells with those of the infusion products of several publicly available clinical trials showed a signature similar to non-responding CLL, DLBCL and MM patients.Thus, abundant viral integrations trigger tonic signalling in CAR High T cells and create cytotoxic but also pre-exhaustion transcriptional signatures in pre-infusion products, jeopardizing longterm treatment.CAR High T cells are therefore the consequence of unsupervised, high viral integrations of CAR transgene, so investigating potentially weaker promoters to insert CAR vectors in or devising methods to manage the number of viral integrations could be the focus of future research [70].As explained, CRISPR/Cas9 can substitute stochastic viral transduction of CAR transgene with a more precise, controlled integration.This results in uniform CAR expressions in the product [71].Multiomics approaches have confirmed how CRISPR-mediated CAR integration in the TRAC locus results in the production of CAR T cells with an enhanced antitumor response and lower exhaustion signatures compared to CAR T cells produced using retroviral vectors [72].Unstimulated virus-free CRISPR CAR T cells created by Mueller et al. exhibited lower basal TCR and CAR signalling and cytokine production than retroviral CAR T cells, but their cytokine production rose to a similar or higher level than retroviral CAR T cells after antigen stimulation.Based on the scRNA-seq data, unstimulated CRISPR CAR T cells had almost double the number of memory-like cell clusters and half of the effector-like clusters compared to retroviral CAR T cells.Upon stimulation, memory-like transcriptional signatures of CRISPR CAR T cells remained stable, unlike retroviral CAR T cells that shifted more towards effector signatures [72].These observations, which are due to the decrease in tonic signalling within CAR T cells after TRAC disruption, emphasize the importance of CRISPR/Cas9 in quality CAR T cell production.
The costimulatory domain(s) of CAR T cells introduced in the second generation are needed to properly activate T cells.Even though the costimulatory domains are functionally similar, their CAR T cells have different kinetics and metabolisms in the clinic [73].Bulk and single-cell transcriptional profiling related to 4-1BB and CD28 CAR T cell costimulatory domains were previously explored at rest (unstimulated) and after activation compared to untransduced T cells.The bulk analysis identified most of the differentially expressed genes between 4-1BB and CD28 that were interrogated by scRNAseq [74].While the activation signatures of 4-1BB-CD3ζ (BBζ) and CD28-CD3ζ (28ζ) CAR T cells were similar and still lower than TCR strength, their top activation genes were the same, which indicates a similar response to activation.The gene expression pattern of BBζ included the enrichment of genes and gene sets related to fatty acid metabolism, the IL-21 cytokine axis (important in CD8+ memory T cell formation) and Th1 polarization.The latter is crucial because Th1 cytokines tend to cause CRS in patients.BBζ also displayed tonic signalling in the absence of a ligand.Interestingly, BBζ CARs were found to be enriched in CD8+ central memory cells, while 28ζ CARs had more CD8+ effector and CD4+ central memory cells [74].All these results explain why BBζ CAR T cells exhibit more expansion and persistence in patients, and how co-stimulatory domains impact the transcriptional CAR T cell profiles.
One of the main challenges of CAR T cell therapy is the on-target off-tumour toxicity due to the simultaneous expression of target tumour antigens in normal tissues, which is associated with poor prognosis.scRNA-seq can be employed to estimate toxicity risk in various organs.The toxic potential of anti-mesothelin CAR T cell therapy was previously explored by analysing existing online scRNA-seq data on mesothelin expression in different tissues [75].According to the scRNA-seq datasets, the organs were classified into high or low risk based on the mesothelin expression.Results put myocardial cells at the highest risk followed by respiratory and digestive system cells.The oesophagus, ileum, liver, kidney and bladder had the lowest on-target off-tumour toxicity [75].This methodology is feasible in any type of CAR T cell therapy to evaluate the risk versus reward of the therapy, prepare clinicians for supportive care, and help with establishing novel TAAs [76].To further elucidate the underlying factors for ICANS, another group used scRNA-seq to identify CD19 expression in brain mural cells, which support blood-brain barrier integrity and are a contributing factor to neurotoxicity in anti-CD19 CAR T cell therapy [77].
Several studies are using scRNA-seq to interrogate tumour and CAR T cell transcriptional profiles in clinical settings.The fact that leukaemia, lymphoma and MM tumours are nested in BM along with their CAR T cells makes analysing such cells particularly demanding.Numerous studies are taking advantage of scRNA-seq in CAR T therapy, nonetheless.Tracking the dynamics of BCMA-directed CD4+/CD8+ (2:1 ratio) CAR T cells in patients with refractory primary plasma cell leukaemia before (Day 0), at peak (Day 8) or during remission (Day 15) phases after the infusion was the focus of a study [78].As in previous in vitro studies, scRNA-seq classified CAR T and donor T cells into distinct clusters based on the expression of certain gene sets, and the relative proportion of cell clusters dramatically changed at peak and remission stages.CD8+ GZMB+ T cells constituted most of the peak CAR T (98.1%) and T (89.1%) cells, while CD4+ CAR T cells were virtually non-existent in remission.Remission phase CAR T cells had significant expression levels of ribosomal protein (RP) genes, which play a role in memory T cell formation [79].The analysis of other TME cells upon interaction with CAR T cells would have improved interpretation of the results.
scRNA-seq has been applied to investigate if a correlation between transcriptional signatures and a clinical response exists in LBCL patients during CD19-directed CAR T cell therapy [80].In a 3-month treatment, patients were divided into progressive disease (PD), partial response (PR) and CR groups.Notably, CRs showed CD8+ central memory signatures that included the expression of CCR7, CD27 and SELL.On the other hand, PR/PD groups demonstrated massive expression of CD8+ and CD4+ exhaustion genes.Notably, a small cluster of 254 cells with transcriptional features of the myeloid lineage (SPI1, LILRB4 and SIRPA) and significant cytokine and chemokine genes expression (IL1B and CXCL8) was over-represented in the infusion products of high-grade ICANS patients.Although the signature of these cells was most similar to classical monocytes, the lack of canonical monocyte markers does not completely classify them as monocytes, so their origin and lineage were unknown [80].Validating the association of a certain cell type in CAR T cell products with adverse effects makes for a reliable predictive marker.
Another application of single-cell analysis is the identification of TME-mediated mechanisms of CAR T primary resistance and disease relapse (acquired resistance), which are led from different origins [81,82].The preexistence of CD19 neg clones of leukaemic cells, for instance, which survived and flourished after CART19 therapy, was elucidated by bulk and single-cell transcriptomic assays [83].scRNA-seq can inspect inherent cellular defects in autologous cell therapies or refractory tumour cells.One such defect in tumour cells is the loss of death receptors, such as TNFR and Fas.Singh et al. used CRISPR/Cas9-based genome-wide knockout screen in Nalm6 cells and cultured them with CART19 cells [84].The main finding was that cancer cells that had lossof-function in death receptor signalling molecules (FADD, BID, CASP8, TNFRSF10B) were resistant to CART19 cells.In addition to being resistant to CART19 cells, the interaction of these leukaemic cells with CAR T cells reduced their expansion and cytotoxic cytokine production, increased the expression of exhaustion genes, and led to CART19 dysfunction.Subsequent scRNA-seq analysis on paediatric and adult ALL patients showed that responders had significantly higher death receptor expression than non-responders, and death receptor signature score was correlated with overall survival, which reveals death receptor to be a potential predictive biomarker as well.These results delineate how immune cell defects are not all inherent and can be acquired following tumour inception [84].A recent study used scRNA-seq and scTCR-seq on the TME of relapsed mantle cell lymphoma (MCL) patients after CD19-directed brexucabtagene autoleucel therapy to report the TME transcriptomic alterations and its T cell clonal expansion during remission and relapse [85].They longitudinally sampled and sequenced TME cells that consisted of MCL B cells, non-tumour cells (NK cells, CTLs, monocytes, etc.), and only 12 CAR T cells (excluded from the analysis).The cellular composition of the TME changed over time.The proportion of lymphoid (especially CD4+ and CD8+ CTLs) and myeloid cells significantly decreased and increased respectively after relapse compared to pretreatment and remission TME cells.Relapse CD8+ CTLs were less cytotoxic (low GNLY and KLRD1 expression) and more exhausted (high TIGIT expression).Among exhaustion-associated genes, only TIGIT expression was consistently high in CD4+ and CD8+ CTLs in all samples.LAG3 was also elevated in most patients (only in T and NK cells), but only a small number of CTLs expressed PDCD1, CTLA4 or TIM3.T cell clone size increased during remission but decreased after relapse according to scTCR-seq.In the myeloid lineage, myelocytes and neutrophils of the TME had a marked increase.MDSCs were among the increased cells in relapse with high expression of CLU, VCAN, VSIR and PIM1 activation markers that induce tumour growth and MDSC survival.Finally, MCL cells had a pronounced decrease in CD19, CD20, CD22, CD79 and MHC-II expression after relapse.TIGIT expression in refractory MCL cells remained longitudinally stable but significantly increased in relapse.TIGIT suppresses immune cell function by binding to CD155 of T and NK cells and promotes immune evasion by the TIGIT/CD155/CD226 axis [85].These results paint TIGIT as a potential target to prevent relapse.This was the first report of acquired TIGIT expression in the tumour cells of a haematological malignancy.More studies are needed to clarify the role of TIGIT in this and other cancers.However, it seems that targeting TIGIT can negate tumour-mediated immune cell suppression (e.g., CTLs and NK cells) and prevent relapse.
The above-mentioned transcriptomics data have uncovered much regarding the composition and heterogeneity of TME.One disadvantage of this method is the need to destruct tumour tissue into single cells, losing information on the position of TME cells relative to other cells or non-cellular structures and their intercellular signalling.It also makes the dissociation process difficult when cells, such as brain neurons, are naturally anchored to the tissue.Selected as 2020 Method of the Year by Nature Methods, spatial transcriptomics helps circumvent these issues and provide more data regarding the RNA repertoire of cells [86].The overarching aim of spatial transcriptomics is to count the number of specific transcripts in certain locations in a tissue.It can be utilized in tandem with scRNA-seq as well [87].Xia et al. performed spatial transcriptomics on hot and cold tumours and combined the data with 14 964 single-cell transcriptomes of primary central nervous system lymphoma (PCNSL) patients [88].This integrative analysis annotated cancer cell subclones and immune cells and their spatial distribution to see which tumour cells reshape the TME and create a cold tumour.Unlike cold tumours that were infiltrated by a small number of T cells and had distinct clusters of tumour cells and uneven immune cell dispersion, large numbers of T cells were detected inside and in the margins of hot tumours.Gene ontology enrichment analysis on tumour clusters showed up-regulated gene sets associated with positive and negative immune response in hot and cold tumours, respectively.A closer look at genes related to response to stimulus (termed 'organizers') showed only FKBP5 to be closely related to tumour progression of the patients.By calculating the single-cell gene expression profile of T cell subsets and scoring the spot of each subset in the spatial transcriptome, they found CD8+ T cells to interact the most with other TME cells in hot tumours, and CXCR4-CXCL12 was the most activated signalling axis.In cold tumours, CD99 signalling axis (a potential CAR T cell target) was the most spatially distributed in the TME, and tumour subclones with the highest proliferation capacity received this signalling the most.Cold tumours had built up barriers of cells with tight intercellular junctions, and the FKBP5+ tumour subgroup was key in creating these barriers.CD19+ tumour cells were encapsulated inside these barriers, which explains why CAR T cells have difficulty penetrating solid tumour margins [88].These results could help decide the proper course of therapy based on TME structure.

PROTEOMICS
Although single-cell transcriptomic studies give a wealth of information on CAR T cells, data gathered from transcripts may not provide the complete picture because of rapid mRNA degradation, alternative splicing and posttranslational modifications [89].Indeed, because proteins are biological workhorses, investigating them can give us a deeper understanding of cells.With the help of proteomics, first coined in 1995, it is now possible to catalogue and characterize the proteins within a cell, tissue, organ or organism (proteome) [90,91].The most popular methods for analysing the proteome of a single cell include high-resolution imaging coupled with genetically encoded fluorescent proteins (as employed in the Human Atlas Protein project), antibody-based approaches coupled with flow and mass cytometry, and mass spectrometry.However, these approaches have certain limitations in terms of sensitivity, resolution and multiplexing.These drawbacks make these approaches ineffective in analysing the whole-proteome within a particular cell [92,93].CyTOF (cytometry by time-of-flight) mass cytometry, for instance, has a limited recognition of roughly 50 surface proteins when most human cells express between 500 and 1000 times more surface proteins [94].In CAR T cell studies, single-cell barcode chips (SCBCs) that utilize microarray probes to detect proteins [95], cellular indexing of transcriptomes and epitopes by sequencing (CITE-Seq) that performs RNA sequencing along with surface protein characterization on a single-cell level [96] and mass spectrometry that measures the massto-charge ratio of one or more molecules in/on cells [97] are the most commonly used techniques for characterization and determination of proteins and phenotypes.
Trial data from CD19 CAR T cell therapies in B cell malignancies confirm that tumour antigen downregulation/escape is one of the common mechanisms of CAR T failure [98].Engineering multi-specific (capable of targeting more than one tumour antigen) and polyfunctional (capable of co-producing multiple cytokines/chemokines at the single-cell level) CAR T cells is a potentially effective approach to overcome this obstacle in cancer immunotherapy [14,98,99].The first step in designing multi-specific CAR T cells is to determine targetable molecules on the tumour cell surface.The collective surface proteome (surfaceome) of tumour cells is also pivotal for TME interactions, highlighting the role of surfaceome analysis in antibody-based therapies.Bulk surfaceome analysis of MM patients identified the CCR10 chemokine receptor and the activated conformation of integrin β7 as promising targets for CAR T cell therapy [100,101].Ferguson et al. have recently used a modified mass spectrometry approach to simultaneously capture 530 surface proteins on MM cell lines, among which immunotherapy targets and flow cytometry markers were present [100].By combining the results with online RNA-seq data, proteins with the highest expression in myeloma plasma cells and the lowest in normal cells were identified.CCR10, TXNDC11 and LILRB4 (an immunotherapy target in AML) met these criteria by having marked MM cell expression and low haematopoietic cell expression.Proof-of-principle anti-CCR10 CAR T cells were able to lyse myeloma cells as well.One interesting application of tumour surface protein analysis is characterizing relatively specific but highly abundant antigens for CAR T cell targeting or co-targeting.This approach to increase CAR T cell avidity through high-abundance antigens can be used to make CAR T cells with low CAR density physically interact more with and dwell longer on tumour cells to exert their killing function [102].CD48 was the most highly-expressed protein on myeloma cells in the previous study [100] and, though it was already shown to be an immunotherapeutic target [103], the methodology can be used to explore potential CAR T cell targets.An example would be CD72 as an optimal CAR T cell target for B-ALL malignancies [104].Though these studies were on bulk samples, future experiments can implement single-cell proteomics on different clones (like treatment-resistant or susceptible) of a range of malignancies to unearth potential CAR T cell target antigens.
Polyfunctionality is another key aspect of durable cellular response as evidenced by the most potent tumour-infiltrating lymphocyte (TIL) subsets being polyfunctional [105].Characterizing the cytokine production of CAR T cells permits us to further identify the heterogeneous functions of phenotypically identical immune cells and draw correlations with clinical outcomes or toxicities.By capturing and detecting the cytokine secretion patterns of stimulated CAR T cells on a single-cell level, their functional profiles can be analysed.For instance, CD4+ and CD8+ CAR T cells may be divided into effector (e.g., granzyme B, IFN-γ, perforin, MIP-1α, TNF-α), stimulatory (GM-CSF, IL-2, IL-5, IL-8, IL-12, IL-15, IL-21), chemoattractive (MIP-1β, CCL-11, IP-10, RANTES), regulatory (IL-4, IL-10, IL-13, IL-22, TGF-β1, sCD137) and inflammatory (IL-1β, IL-6, IL-17A, IL-17F, MCP-1, MCP-4) subsets [106,107].Accordingly, singlecell proteomics data demonstrated that NHL patients respond best to CD19 CAR T cells capable of coproducing IFN-γ, IL-8 and/or MIP-1α [108].Moreover, objective responses were also detectable in NHL patients with CD8+ CAR T cells expressing granzyme B, and CD4+ CAR T cells co-expressing IL-17A and IL-8, and/or IL-5 and IL-8 [108].The use of scRNA-seq to study CAR T cells activation functions was first reported by Xhangolli et al., who co-cultured CD19-CD28/4-1BBζ CAR T cells with B cell lymphoma Raji cell lines and correlated the scRNA-seq results to single-cell cytotoxicity and single-cell cytokine assay [109].Although the activated populations of CAR T cells secreted a wide range of cytokines instead of cytokines specific to cytotoxicity (like granzymes), they were all correlated with a cytotoxic response.This shows that CAR T cells' cytotoxic function is a heterogeneous response and not bound to a few certain cytokines.They found that both CD4+ and CD8+ CAR T cells had equal killing effectiveness regardless of their differentiation status due to equal expression levels of effector molecules and cytokines.Differentiation status (naïve, central memory, effector memory and effector CAR T cells) was independent of cytokine production.Intriguingly, mapping transcriptome data to phenotypes showed that the most potent CAR T cells were cells that exhibited a mixed Th1/Th2 polarization state, which a bulk analysis would have failed to discern.Proteomics analysis can thus be used in conjunction with scRNA-seq as a forward approach to identify key players linked to a particular phenotype [109].In line with these findings, a single-cell multi-omics study on a retrospective cohort of patients receiving CAR T cells (anti-CD19-4-1BBζ) indicated that relapse in CD19+ B-ALL patients is often associated with a decline in the expression of Th2 CAR T cell cytokines (IL4, IL5, IL13) [110].Multi-specific CAR T cells are advantageous in that they can still spot their target after one antigen loss.Single-cell proteomics data from a phase I/II clinical trial (NCT04186520) measured the polyfunctional strength index (PSI) of bispecific CD19/CD20 CAR T cells (LV20.19) in B-ALL patients and revealed that these cells exhibited heightened levels of polyfunctionality compared to mono-specific CD19 CAR T cells [107].The PSI value of CD20-activated LV20.19 was similar to that of CD19-activated CAR T cells, which suggests that such cells maintain their cytotoxicity against CD19 neg tumours.This study, too, found higher TGF-β and IL-17A (CD4+ CAR T cell cytokines) to be most correlated with clinical response.Collectively, according to single-cell proteomics and multi-omics studies, the presence of polyfunctional CAR T cells able to co-produce Th1 and, more importantly, Th2 cytokines contributes to a partial or CR in patients with B-cell malignancies.However, other intrinsic and extrinsic features of CAR T cells can also determine the clinical efficacy and outcome of CAR T cell therapy.
Apart from polyfunctionality, other phenotypical characteristics of CAR T cells, including persistence and memory phenotype, are known to influence clinical outcomes [4,111,112].Accordingly, a CITE-seq analysis in a retrospective study demonstrated that a low abundance of central memory (T CM ) or stem cell memory (T SCM ) CAR T cells (CD62L + CCR7 + CD45RO+), rather than activation spectra (CD69, CD38 and HLA-DR expression) or inhibitory signatures (PD-1, CTLA-4, LAG-3 and TIGIT expression), in the pre-infusion product correlates with CD19 positive B-ALL relapse [110].Naïve T cell (T N ), T CM and T SCM proportions are reportedly directly associated with CAR T cell persistence beyond 6 months [113].An initial scRNA-seq of FACS-sorted, pre-manufactured T cells of 71 B-ALL patients delineated the critical role of TCF7 and LEF1 transcription factorencoding genes in maintaining desirable early memory T cell phenotypes.The most differentially upregulated genes in poor prognosis patients were the IFN I pathway genes (RSAD2, IRF7, MX1, etc.).Subsequently, paired CITE-seq and scATAC-seq on the T cells of six of these patients found high TCF7/LEF1 motif chromatin accessibility associated with the naïve and memory T cells, PRDM1 with T EFF and T EM cells, and TBX21 with CD8+ T EM and T EFF cells [113].This study used single-cell analysis to illustrate which regulons in donor T cells are linked to certain T cell subsets and clinical outcomes.More recently, the follow-up results of a successful phase I trial of CAR T cell therapy in 2010 by Melenhorst and colleagues recounted how CAR T cells persisted in two CLL patients with sustained remission for more than a decade.In this work, researchers used CyTOF longitudinally over a 10-year period to characterize CAR T cell phenotypes and protein expression.In patient Number 1, CD8+ GZMB and CD4+ CAR T cells were the initial responders and were prevalent for only a few months.CD4 + Ki67 High cells then dominated the CAR T cell population by constituting at least 97% of CAR T cells by Year 1.4.In the second patient, this transition from CD8+ CAR T cells was delayed by Year 4.7, and by Year 5, almost half of the CAR T cells were CD8+ GZMK, CD8+ GZMB, CD4+ and CD4/CD8 double-negative (DN) Helios High CAR T cell populations.By Year 7.2, CD4 + Ki67 High constituted 97.6% of CAR T cells.A combined CITE-seq/scTCR-seq analysis showed the DN Helios High CAR T cells to be γδ CAR T cells as evident from the expression of TCRγδ, lack of TCRαβ clonotype, and TRDV1 and TRGV4 transcriptional expression [5].Moreover, proteomics and transcriptomics data showed that CD4 + Ki67 High CAR T cells also have a distinctive profile.They expressed activation markers CD38, HLA-DR and CD95, transcription factors EOMES and TOX, checkpoint markers CTLA-4, LAG-3 and TIGIT and memory markers CD27 and CCR7 [5].The comparison of CD4+ CAR-non-expressing T cells with CD4+ CAR T cells identified 645 differentially expressed genes, and the CD4+ CAR T cells had a significant enrichment for genes related to TCR and T cell signalling and activation, oxidative phosphorylation, and so forth, which explains the cytotoxic traits of these cells and the patients' favourable outcome.These single-cell analysis data give invaluable information regarding long-term remissions after CAR T cell therapy.
Several other proteomics and multi-omics experiments have been conducted to improve our understanding of another recurring theme in CAR T cell resistance: CAR T cell exhaustion.A multi-omics research on an in vitro dysfunction model of CD8+ anti-mesothelin CAR T cell showed that upon exhaustion and dysfunction, several NK cell-related proteins, including the inhibitory receptors (KLRB1, TIGIT, NKG2A, PD-1), CD56 and granulysin, are upregulated [114].Continuous antigen exposure over a maximum of 35 days resulted in the surface expression of exhaustion markers as well as the internalization of surface CAR and repressed cytotoxicity (both restored after 1 day of rest in fresh media and IL-15).The progressive exhaustion and NK phenotype results were corroborated by bulk RNA-seq and ATACseq: The temporal gene expression analysis revealed a gradual expression of exhaustion markers and NK cell receptor genes, which displayed chromatin opening as well.RNA-seq of the TF genes listed the upregulated (EGR1, ID3, SOX4, RBPJ) and downregulated (KLF2, BCL6, LEF1) expressions.scRNA-seq was able to exclusively identify select genes (RGAP3, DUSP4, CSF1) not yet implicated in exhaustion.A final mass and flow cytometric profiling of the continuously stimulated cells suggested that CAR T cell dysfunction is highly associated with a CD8+ T to NK-like T transition, and blocking this pathway by targeting related TFs could contribute to improved CAR T cell efficacy in clinical settings [114].Jackson et al. explored the transcriptomic and phenotypic profiles of post-infusion CART19 cells versus pre-infusion ones in NHL patients [115].First, the average postinfusion CD8+ CAR T cell proportions in the patients increased from 52% to 87%, and the total effectorassociated TFs (PRDM1 and EOMES), exhaustionassociated TFs (TOX, TOX2, NR4A2, AP-1 family), cytotoxic effector molecules (GZMB, PRF1, GZMK, CCL5) and exhaustion (CTLA4, LAG3, HAVCR2, PDCD1, VSIR, TIGIT) genes were elevated in the clusters as well.These were in line with surface expression analysis where, from Days 0 to 14, global surface expression of CD45RA, CD127 (IL-7R), CD62L and CD25 (naïve and memory marker) was surpassed by CD45RO, CD69, CD57 and PD-1.A global increase in CD127, CD25 and CD197 memory markers was observed from Days 14 to 30.PD-1 and TIGIT surface expression were the highest among exhaustion markers on both CD4+ and CD8+ CART19 cells; however, TIGIT was the most sustained during 30 days.Unsurprisingly, CD8+ CART19 cells of non-responders were less proliferative and more dysfunctional than responders, but the difference for CD4+ CART19 cells was not significant.In non-responder CD8+ CART19 cells, TIGIT had the highest upregulation from pre-infusion among exhaustion genes both globally and in predominant clusters.The population of these cells was dominated by T EM, and they transitioned to T eff quicker than in responders which might indicate early unnecessary CAR overstimulation prior to infusion.Although it would have been prudent to profile the exhaustion markers of tumour-isolated CART19 cells as well, these findings established TIGIT as a master regulator of CAR T cell exhaustion [115] similar to previous studies [85], proposing CAR T cell therapy in combination with TIGIT blockade as a potential approach for the treatment of NHL patients.
In addition to CAR T cell phenotypes, CRS and ICANS are two significant side effects associated with CAR T cell therapy [116].Single-cell proteomic profiling revealed that post-infusion CD4 + Helios + CAR T cells, which manifest hallmark characteristics of regulatory T cells, are associated with progressive disease and less severe ICANS in LBCL patients.In contrast, expression of CD57 and T-bet was shown to be linked to durable disease control [117].Furthermore, a serum profiling study discovered that IL-18 is also associated with ICANS during CD19 CAR T cell therapy [118].Their findings also identified FMS-like tyrosine kinase 3 (FLT3) and mast cell immunoglobulin-like receptor 1 (MILR1) as two pre-infusion predictive biomarkers of severe CRS.Still, additional single-cell proteomics data are required to successfully and more accurately predict clinically significant CRS and ICANS to alleviate patient conditions.
In conclusion, bulk and single-cell proteomics data help scientists understand the relationship between in vitro function profiles of CAR T cells and therapeutic outcomes [119].Moreover, these data can be useful in engineering and designing polyfunctional, persistent, memory-like, exhaustion-resistant CAR T cells (Table 3).However, single-cell proteomics data are limited in CAR T cell therapy in solid tumours.Hence, additional fundamental single-cell proteomics research is required to better understand the pathways and mechanisms behind CAR T cell dysfunction in solid tumours and tumour relapse following CAR T cell therapy.

THE TRADE-OFFS OF SINGLE-CELL SEQUENCING
Single-cell technologies can help address manufacturing challenges of CAR T cells.With the aid of cell engineering tools, they can optimize CAR T cell development and personalize patient care [120].One advantage is identifying novel targets that reduce the chance of ontarget off-tumour toxicity.Single-cell transcriptomics and/or proteomics are being used to annotate candidate protein expression in tumours and healthy tissues to validate potential targets that are most associated with a certain cancer [121].Analysing the omics of persistent tumour subclones in a R/R patient allows us to produce CAR T cells specifically against said clones.A single-cell atlas of infused CAR T products can also be assembled to identify similarities and differences between the products and their correlation with clinical outcomes [122].Single-cell analyses can identify robust markers of CAR T persistence as well.In 10 successful cases of R/R B-ALL treatment, distinct transcriptional signatures were seen in long-lived CAR T cells 5 years after therapy.This could help with patient selection and optimizing manufacturing protocols that produce persistent CAR T cells [123].Long-term manufacturing of CAR Ts is shown to abolish T CM transcriptional profiles.Manufacturing protocols, CAR design and architecture, and so forth alter the end-product [124].Single-cell analyses can help compare protocols to single out the most suitable ones.One potential use would be to use single-cell transcriptomics or epigenomics analyses to calculate the exhaustion level of CAR T cell products [68].T cell exhaustion directly impacts treatment outcomes.Screening products with low signatures of activation genes and tonic signalling as well as low exhaustion profiles is possible with single-cell technologies.The products can also be screened for having larger populations of T CM cells [74].CAR T products can also be assessed based on their functional profile.An issue with CAR T cell therapy is the general lack of knowledge regarding the content of these live drugs.Overall, multi-omics tools can calculate the prevalence of each T cell phenotype because some patients might respond more favourably to certain phenotypes [108].
Though single-cell analysis is deepening our knowledge of heterogeneous cell populations, it still suffers from various technical and computational pitfalls, especially compared to bulk analyses.Cell dissociation from tissue matrix can be problematic in some organs.The cells must first be separated by harsh methods, such as enzymatic digestion or mechanical separation, which are detrimental to the cells.Such methods can cause membrane rupture or cell stress.Stress is shown to trigger transcriptional changes that bias cell phenotypes, serving as a confounding factor [125].This is exacerbated in some organ-specific cells that are refractory to enzymatic dissociation, causing them to be under-represented in the analysis.In scRNA-seq, some RNA transcripts might become lost during reverse transcription because of low copy number, which is known as a dropout event.This results in missing some transcript in one cell while they are expressed in another [126].Stricter protocols are required for single-cell analysis library preparation.The massive amount of generated data from novel methods acts as a double-edged sword, making data handling laborious [127].The number of analysed cells is conversely related to coverage.Including more cells results in shallower coverage and costlier analysis and while it does allow the recognition of rare subpopulations, less prevalent transcriptional profiles might be lost [128].
Sample handling is also crucial to the analysis outcome.This is especially important in single-cell analyses where data are sensitive to individual cell profile alterations.Late sampling or processing samples in over 2 h is shown to introduce gene expression and chromatin accessibility artefacts [129,130].The transcriptomics and proteomics data from same cells under similar conditions lack correlation at times.Biologically, this is mostly due to posttranslational modifications, RNA silencing or different half-lives of proteins and RNA that affect cell physiology, and transcriptomics analyses are blind to them.A mechanistic understanding of both mRNA and protein is needed to interpret such data as best as possible [131].
The scarce nature of datasets and the massive amount of data make the computational aspect challenging.The accurate analysis of single-cell data requires extensive [117] ALL tumour cells N/A scRNA-seq and CyTOF • multi-omics analysis of bone marrow of non-responding patients showed pre-existence of leukaemic subclones with stem-cell like markers and low antigen presentation [10] Abbreviations: ALL, acute lymphoblastic leukaemia; HD, healthy donor; LBCL, large B cell lymphoma; LV, lentiviral; MIP-1α, macrophage inflammatory protein-1 alpha; NHL, non-Hodgkin lymphoma; SCBC, single-cell barcode chip.
biological expertise and an adept data analyst, and standardized pipelines are needed for the translation of data analysis [132].

CONCLUSIONS
Although patients with advanced haematological conditions can benefit from CAR T cell therapy certain drawbacks in CAR T cell therapy, including but not limited to CAR T cell exhaustion and short-term persistence of infused CAR T cells, make CAR T cells less effective in the treatment of patients with solid tumours [1].Additional research is therefore required to address these obstacles and improve the safety and efficacy of CAR T cells in cancer patient.In this review, we discussed the significance of single-cell analysis in CAR T cell therapy at several single-cell levels, including genomics, epigenomics, transcriptomics and proteomics.Compared to bulk sequencing data, the development and leveraging of single-cell sequencing technologies are accelerating, giving us the capacity to address challenges in CAR T cell treatment in unprecedented depth.
As was previously noted, CAR T cell persistence and effector function are key determinants in clinical outcomes and, given that CAR integration at certain sites might impact CAR T cell persistence and effector function, scDNA-seq and scATAC-seq can be utilized to identify the optimum site for CAR transgene integration.Moreover, data gathered from single-cell analysis, particularly scRNA-seq data from CAR T cells can be used to map and identify genes associated with a favourable clinical response.For instance, TCR tracing in conjunction with other single-cell omics technologies can be employed to identify dominant clones and subclones of CAR T cells with enhanced antitumor ability and to identify gene signatures associated with potent CAR T cells [38,40].Single-cell data from these dominant clones can be used to engineer the next generations of CAR T cells to treat non-responding patients.However, scRNA-seq has a few drawbacks, such as the blurred occasional lack of correlation between gene expression and protein expression, which we briefly discussed.ScRNA-seq has other limitations, such as biases in transcript coverage, suboptimal capture efficiency and sequencing coverage.As a consequence, scRNA-seq data typically exhibit higher levels of technical noise compared to bulk RNAseq data [133].Even the most sensitive scRNA-seq protocols frequently encounter a common issue known as 'dropout events', where certain specific transcripts cannot be reliably detected [134].Moreover, scRNA-seq experiments often yield data of subpar quality due to factors like cell breakage, cell death, or the presence of mixed cell populations [135].These low-quality cells can significantly impede downstream data analysis and potentially result in data misinterpretation.Therefore, it is imperative to perform quality control on scRNA-seq data to identify and eliminate these low-quality cells.Single-cell proteomics data can aid to some extent in overcoming this challenge.Preliminary proteomics findings discovered that the Th2 profile is a decisive factor in CAR T cells with enhanced antitumor capability, which could not be easily discovered with other omics technologies [105,106].Still, limited single-cell proteomics data for CAR T cells in solid tumours are available and further single-cell proteomics investigations, both in T cells and CAR T cells, are required to uncover the pathways and mechanisms underlying CAR T cell dysfunction in solid tumours and tumour relapse following CAR T cell therapy.
Future studies will also explore the possibility to adapt single-cell radiomics such as using single-cell fluorescent probes alone or in combination with other single-cell modalities (e.g., metabolomics, epigenomics, etc.) to investigate whether radiomics or serum metabolomic/ epigenomics signatures could be predictive or prognostic factors in the follow-up of patients receiving CAR T cell therapy.Altogether, single-cell omics can be used not only to determine and discover cellular genotypes, phenotypes or metabolites correlated with the best clinical response to CAR T cell therapy, but also to define the relationship between pre-and post-infusion CAR T cells to design effective, polyfunctional, persistent, memory-like and exhaustion-resistant CAR T cells.Moreover, these data can fundamentally be used to uncover the mechanisms underlying CAR T cell toxicity, exhaustion, therapy resistance or relapse and patient selection for CAR T cell therapy.Finally, the data generated by single-cell sequencing can be used to identify specific subtypes of CAR T cells possessing superior antitumor activity.This information can then be used to develop new generations of CAR T cell therapies that more effectively target tumour cells.
AUTHOR CONTRIBUTIONS Sasan Ghaffari and Hamid Reza Mirzaei contributed to the concept of the review.Sasan Ghaffari, Mahshid Saleh, Behnia Akbari and Faezeh Ramezani were responsible for the reference selection and drafting of the manuscript.Hamid Reza Mirzaei contributed to the critical review of the manuscript.Sasan Ghaffari drew the figure and amalgamated the outlines.All authors read and approved the final manuscript.
levels of single-cell analysis.These analyses can be conducted on various tumours, CAR-T and T cells, circulating tumour DNA and every major organ.Genomics analyses can track mutations in heterogeneous tumours (a).scDNA-seq can show which tumour mutations are associated with poor clinical outcome, relapse or CAR-T resistance (b).TCR sequencing allows tracking the lineage of various CAR-T cell clones (c).Epigenomics can show what epigenetic alterations and transcription factors are associated with CAR-T exhaustion or long-term CAR-T cells (d), or how CAR gene insertion epigenetically alters CAR-T cells (e).We can distinguish CAR-T cell subtypes or tumour populations in distinct clusters based on their transcriptional signatures (f).These profiles allow us to identify exhaustion-related gene expressions, how CAR integration manipulates gene expression, correlate expressions with clinical outcomes or predict on-target off-tumour toxicity (g).Transcriptomics can delineate TME mechanisms of CAR-T cell resistance or relapse as well (h).Surfaceome tumour analysis can depict potential CAR-T cell targets with a low chance for on-target off-tumour (i).Proteomics analysis of CAR-T cells can determine their phenotypes and cytokine production profiles (j) for the identification or drawing correlations with CAR-T exhaustion, toxicity, relapse, or clinical outcome (k).The Figure was partly generated using Servier Medical Art, provided by Servier, licensed under a Creative Commons Attribution 3.0 unported licence.
Pyrosequencing and bisulfite genomic sequencing • DNA methylation was detected in the CpG of many TCR signalling regulatory genes in transduced, pre-infusion CAR T cells.

T
A B L E 2 A summary of the preclinical and clinical transcriptomic studies in CAR T cells.
RNA and ATAC-seq, scRNA-seq • Upon CAR T cell exhaustion, CAR T cells show high expression of NK-related proteins and an NK-like phenotype.• High levels of NK-like CD19-CAR T cells postinfusion predict treatment failure [114] NHL CAR T cells, pre-and postinfusion LV αCD19-4-1BB-CD3ζMulti-omics, 10Â feature profiling•TIGIT expression is associated with high levels of nonproliferative, highly differentiated, and exhausted CD4+ and CD8+ CAR T cells in NHL patients.• CD8+ CAR T cells of nonresponders were significantly more exhausted than responders [115] LBCL CAR T cells, post-infusion LV αCD19-CD28-CD3ζ CyTOF and CITE-seq • Presence of CD4 + Helios+ CAR T cells on day 7 post infusion can predict less severe neurotoxicity and disease progression while the presence of CD57 + T-bet+ cells is associated with complete response in patients.

Attribution 3 .
0 unported licence.The authors thank the efforts of Susan Moradinasab for her transcriptomics literature search.
T A B L E 3 A summary of proteomics studies in CAR T cells.