Integrative Bulk and Single-Cell Profiling of Premanufacture T-Cell Populations Reveals Factors Mediating Long-Term Persistence of CAR T-Cell Therapy

Published OnlineFirst April 5, 2021; DOI: 10.1158/2159-8290.CD-20-1677

Research Article

Integrative Bulk and Single-Cell Profiling of Premanufacture T-cell Populations Reveals Factors Mediating Long-Term Persistence of CAR T-cell Therapy

Gregory M. Chen1, Changya Chen2,3, Rajat K. Das2, Peng Gao2, Chia-Hui Chen2, Shovik Bandyopadhyay4, Yang-Yang Ding2,5, Yasin Uzun2,3, Wenbao Yu2, Qin Zhu1, Regina M. Myers2, Stephan A. Grupp2,5, David M. Barrett2,5, and Kai Tan2,3,5

abstract The adoptive transfer of chimeric antigen receptor (CAR) T cells represents a breakthrough in clinical oncology, yet both between- and within-patient differences in autologously derived T cells are a major contributor to therapy failure. To interrogate the molecular determinants of clinical CAR T-cell persistence, we extensively characterized the premanufacture T cells of 71 patients with B-cell malignancies on trial to receive anti-CD19 CAR T-cell therapy. We performed RNA-sequencing analysis on sorted T-cell subsets from all 71 patients, followed by paired Cellular Indexing of Transcriptomes and Epitopes (CITE) sequencing and single-cell assay for transposase-accessible chromatin sequencing (scATAC-seq) on T cells from six of these patients. We found that chronic IFN signaling regulated by IRF7 was associated with poor CAR T-cell persistence across T-cell subsets, and that the TCF7 regulon not only associates with the favorable naïve T-cell state, but is maintained in effector T cells among patients with long-term CAR T-cell persistence. These findings provide key insights into the underlying molecular determinants of clinical CAR T-cell function.

Significance: To improve clinical outcomes for CAR T-cell therapy, there is a need to understand the molecular determinants of CAR T-cell persistence. These data represent the largest clinically annotated molecular atlas in CAR T-cell therapy to date, and significantly advance our understanding of the mechanisms underlying therapeutic efficacy.

INTRODUCTION CAR T cells from the bulk population of T cells extracted from the patient, consisting of a mixture of CD4+ and CD8+ T cells Chimeric antigen receptor (CAR) T-cell therapy has been a across naïve, memory, and effector lineages. Clinical and pre- breakthrough in cancer therapy, yet failure to achieve long-term clinical studies have led to a growing understanding of a positive CAR T-cell persistence remains a major barrier to sustained role of naïve and memory T-cell subsets in CAR T-cell efficacy, remission in many patients. Although complete response rates in suggesting that cellular stemness and memory formation are pediatric B-cell acute lymphoblastic leukemia (B-ALL) are high, critical to maintaining long-term remission (4–6). These stud- differences in apheresis T-cell products have been shown to play a ies have provided foundational insight into the composition of critical role in determining the duration of antitumor CAR T-cell favorable T-cell phenotypes, yet a more comprehensive under- function (1). Recent studies have provided strong evidence that standing has been hindered by limited sample sizes, confound- differences in apheresed T cells are a major determinant in thera- ing by T-cell subtype, and lack of deeper molecular analyses. peutic failure of CAR T-cell therapy in adult malignancies such To address these challenges, we developed a subtype- as chronic lymphocytic leukemia (CLL) and multiple myeloma specific transcriptomic atlas of premanufacture T cells from (2, 3), suggesting that a deeper understanding of intrinsic T-cell 71 patients on trial to receive anti-CD19 CAR T-cell therapy. function is essential for improving existing and future CAR T-cell A crucial component of our experimental design was to sort therapies for both pediatric and adult cancers. T-cell subsets prior to RNA sequencing (RNA-seq), allowing At the core of this challenge is the heterogeneous nature of us to directly account for the confounding effect of T-cell autologous T cells that form the starting material for CAR T-cell subset composition and identify clinically associated molec- therapy. Most CAR T-cell therapy trials to date have engineered ular pathways that act within T-cell subsets. We developed a web server for visualization of genes and regulons from 1Graduate Group in Genomics and Computational Biology, University of our transcriptomic atlas, which can be accessed at https:// Pennsylvania, Philadelphia, Pennsylvania. 2Center for Childhood Cancer tanlab4generegulation.shinyapps.io/Tcell_Atlas/. On the basis Research, The Children’s Hospital of Philadelphia, Philadelphia, Pennsyl- of our findings regarding IFN signaling andTCF7 expression vania. 3Department of Biomedical and Health Informatics, The Children’s Hospital of Philadelphia, Philadelphia, Pennsylvania. 4Graduate Group in in effector T-cell subtypes, we performed integrative Cellular Cellular and Molecular Biology, University of Pennsylvania, Philadelphia, Indexing of Transcriptomes and Epitopes sequencing (CITE- Pennsylvania. 5Department of Pediatrics, University of Pennsylvania, seq) and single-cell assay for transposase-accessible chromatin Philadelphia, Pennsylvania. sequencing (scATAC-seq) on six of the patients in the study, Note: Supplementary data for this article are available at Cancer Discovery allowing us to more deeply characterize the interplay and Online (http://cancerdiscovery.aacrjournals.org/). regulation of these clinically associated biological pathways. G.M. Chen, C. Chen, and R.K. Das contributed equally to this article. Corresponding Authors: Kai Tan, The Children’s Hospital of Philadelphia, 3501 Civic Center Boulevard, Philadelphia, PA 19104. Phone: 267-425-0050; RESULTS E-mail: [email protected]; and David M. Barrett, Tmunity Therapeutics, 3020 Naïve and Early Memory T-cell Composition Market Street, Suite 535, Philadelphia, PA 19104. Phone: 215-966-1600; E-mail: [email protected] Predicts Clinical CAR T-cell Persistence Cancer Discov 2021;11:1–14 We identified 71 children or young adults who were enrolled doi: 10.1158/2159-8290.CD-20-1677 to receive anti-CD19 CAR T-cell therapy at the Children’s Hos- ©2021 American Association for Cancer Research pital of Philadelphia (Supplementary Table S1). The mean age

September 2021 CANCER DISCOVERY | OF2

RESEARCH ARTICLE Chen et al.

at enrollment was 11.7 [95% confidence interval (CI), 10.1–13.3], TEFF and TEM Highly Express Genes for T-cell and there was a balanced ratio of female and male patients Activation and Proliferation, but Persistence [47.9% (36.0–60.0) female]. Seventy patients had relapsed or May Be Limited by Increased Apoptosis refractory B-ALL, and one was a young adult with Hodgkin Because the composition of naïve, memory, and effector lymphoma. CAR T-cell manufacture was successful in 65 of the T-cell lineages was robustly associated with clinical CAR 71 patients. The primary clinical endpoint for functional CAR T-cell persistence, we next sought to investigate the biological T-cell persistence was duration of B-cell aplasia (BCA), consist- pathways defining these T-cell subtypes among our patient ent with prior studies which have established that BCA is a sen- cohort. Our bulk RNA-seq data on sorted patient T cells sitive marker of functional anti-CD19 CAR T-cell persistence in captured the functional continuum of T-cell differentiation, pediatric populations (1, 7). Long-term CAR T-cell persistence from TN, TSCM, TCM, TEM, and TEFF subtypes (Fig. 1C). At a was defined as BCA≥ 6 months, whereas short-term or failed global transcriptomic level, between-subtype differences in persistence was defined as BCA< 6 months. Duration of BCA T cells were the dominant effect; T cells of the same subtype was considered to be right-censored if the patient had ongoing clustered strongly between patients, whereas T cells derived BCA at the most recent follow-up, proceeded to bone marrow from individual patients generally separated by subtype. TN transplant, died of other causes, or had CD19− relapse with con- and TSCM shared similar transcriptomic profiles and were tinued BCA. In an intention-to-treat manner, the median dura- largely indistinguishable at the transcriptomic level, whereas tion of CAR T-cell persistence among patients in our cohort was TCM, TEM, and TEFF separated strongly. 11 months, similar to event-free survival durations reported in Differential expression analysis with Linear Models for recent clinical trials in B-ALL (8). Microarray Data (Limma; refs. 10, 11) revealed 6,953 differen- To assess the composition and molecular pathways of pre- tially expressed genes associated with T-cell type (FDR < 0.05; manufacture T cells, we curated a biobank of T cells from these Supplementary Table S2). Pathway analysis of the differen- 71 patients (Fig. 1A). These T cells were aliquoted at time of tially expressed genes revealed an enrichment in lymphocyte clinical leukapheresis and represent the populations of the activation, cytokine signaling, and leukocyte migration path- T cells used for CAR T-cell manufacture. We used FACS to sort ways (P = 1.93e-34, 1.49e-23, 2.54e-21; Fig. 1D); single-sample T cells into five T-cell subsets: naïve (TN), stem cell memory Gene Set Enrichment Analysis (ssGSEA; ref. 12) revealed that (TSCM), central memory (TCM), effector memory (TEM), and effec- these pathways were enriched among the effector T-cell line- tor (TEFF; Supplementary Fig. S1A). We performed RNA-seq ages (Supplementary Fig. S3A). on the five sorted T-cell populations for each of the patients, We found that expression of lymphocyte proliferation path- producing an atlas of 355 transcriptomic profiles representing ways increased along a gradient toward effector T cells (P = the T-cell composition across pediatric premanufacture T cells. 2.5e-15, Fig. 1D and E), associated with higher Ki-67 expres- We found that higher proportions of TN, TSCM, and TCM sion in TEM and TEFF cells (Supplementary Fig. S3A). Effector were associated with clinical CAR T-cell persistence beyond 6 T cells exhibited extensive metabolic reprogramming sup- months [false discovery rate (FDR)–adjusted P = 3.7e-3, 1.3e-2, portive of a functionally activated state. Expression of LDHA, 1.0e-3], and lower proportions of TEM and TEFF were associated GLUT1, and GAPDH was upregulated among effector T cells, with clinical CAR T-cell persistence beyond 6 months (FDR- indicating a metabolic shift to aerobic glycolysis, and path- adjusted P = 4.0e-2, P = 1.1e-3; Fig. 1B). Combined proportions ways involving biosynthesis of protein, DNA, and lipid were of TN, TSCM, TCM, and TEM were associated with clinical CAR enriched, suggesting biosynthetic support for cellular division T-cell persistence, with greatest discrimination observed when (Supplementary Fig. S3B). Counteracting these pathways, we combining TN, TSCM, and TCM proportions (log-rank test P = found that apoptotic pathways, including both intrinsic and 6.2e-3 to 1.4e-2; Supplementary Fig. S1B and S1C), suggest- extrinsic apoptotic signaling, were strongly enriched among ing a general and robust trend for a role of naïve and memory genes expressed higher in the effector lineages (P = 2.6e- subsets as a prognostic marker for clinical CAR T-cell function. 12, Fig. 1F; Supplementary Fig. S3C and S3D). Upregulated We performed CIBERSORTx (9) analysis to estimate the expression of cytotoxic markers, cytokines, and inhibitory relative proportion in CD4+ and CD8+ T cells in each T-cell receptors among TEM and TEFF cells suggests that these sub- subset. To robustly estimate deconvolution proportions, we sets likely represent functional, endogenously activated T-cell ran the algorithm using bulk and single-cell references, find- phenotypes (Supplementary Fig. S4A and S4B). Together, ing that estimated proportions were robust and broadly con- these data suggest that although the more differentiated T cordant with naïve, memory, and activated T-cell phenotypes cells exhibited a proliferative phenotype at baseline, their prolif- based on reference annotations (Supplementary Fig. S2A erative ability may be inherently limited by apoptotic pathways and S2B). We found that TN, TSCM, and TCM subsets were that act as a barrier to the years-long cellular survival and pro- + predominantly CD4 , whereas TEM and TEFF subsets were liferation necessary for long-term success of CAR T-cell therapy. predominantly CD8+ (Supplementary Fig. S2C). Using the deconvoluted data, we found that a greater proportion of + Network Analysis Reveals Critical Roles of TCF7 CD8 TN and TSCM cells was positively associated with clinical CAR T-cell persistence (FDR-adjusted P = 3.1e-2, 4.1e-2; and LEF1 in Maintaining Naïve and Early Memory Supplementary Fig. S2D). This trend was observed both in T-cell States our deconvolution estimates and in differential expression Next, we sought to investigate the transcriptional regulators of CD8 genes (Supplementary Fig. S2E). These data suggest that act to maintain the naïve and early memory phenotypes a key role of naïve T cells, particularly in the CD8+ compart- associated with long-term CAR T-cell persistence. We sought to ment, in contributing to CAR T-cell efficacy. construct a robust transcriptional regulatory network (TRN)

OF3 | CANCER DISCOVERY September 2021 AACRJournals.org

Bulk and Single-Cell Factors Mediating CAR T Persistence RESEARCH ARTICLE

A Premanufacture T cells

TSCM

TCM

TEM

TEFF FACS sorting (71 patients) T-cell subtype RNA-seq

CAR T cells

71 patients enrolled on an anti-CD19 CAR T-cell therapy trial Leukapheresed T cells Single-cell suspension (6 patients) Paired CITE-seq + scATAC-seq

B 0.20 0.3 0.6 P = 3.7e-3 P = 1.3e-2 P = 1.0e-3 P = 4.0e-2 0.20 P = 1.1e-3 0.20

0.15 0.15 N

SCM CM 0.15 EM 0.2 EFF 0.4 0.10 0.10 0.10

Proportion T 0.2 0.1 Proportion T 0.05 Proportion T Proportion T Proportion T 0.05 0.05

0.0 0.00 0.00 0.0 0.00 < 6 months ≥ 6 months <6 months ≥ 6 months <6 months ≥ 6 months <6 months ≥ 6 months <6 months ≥ 6 months CDCAR T-cell persistence Lymphocyte activation Cytokine-mediated signaling pathway Regulation of cytokine production Leukocyte migration PT6 PT60 PT62 PT6 Adaptive immune response PT61 PT31 PT60PT46PT62 PT49 a T Positive regulation of hydrolase activity PT51 PT45 PT42 N PT54 PT56 PT70 PT45PT61PT51 PT31PT35 PT67PT70 Cellular response to tumor necrosis factor PT54 PT57 PT48 PT52 PT40 PT34 a T PT46 PT43PT43PT33 SCM Alpha-beta T-cell activation PT8 PT48 PT47 PT52 PT55 PT44PT44 PT55 PT59PT68 PT47 PT11 a T Myeloid leukocyte activation PT10PT2 PT37 PT41 PT66 CM PT9 PT34 PT66PT71PT71 PT11 PT49 PT68 PT4 Positive regulation of programmed cell death PT67 PT58 PT64PT53 PT58 PT33 PT65 a T PT38 EM Apoptotic signaling pathway PT53 PT38 PT65 PT41 PT40PT42PT35 PT26PT63 PT50 PT37 B-cell activation PT28PT28 PT63 PT64 a T PT50 PT39 PT56 EFF PT15 PT18 IL12 pathway PT27PT23 PT26 PT25 PT36 PT23 PT2 PT3 PT35 Regulation of IκB kinase/NFκB signaling PT25 PT14 PT16 PT13 PT20 PT39 PT12 PT12PT16 PT20 PT36 PT29 PT57 Hemostasis PT7 PT21 PT22 PT17 PT22PT21 PT19 PT19 PT17 PT18 PT8 T-cell costimulation PT7 PT14 PT3PT4 PT59 Cell adhesion molecules PT5 PT27PT9 PT1PT1 PT10 PT24PT24 PT29 Positive regulation of proteolysis PT3 PT9 PT19 Positive regulation of defense response PT69 PT23 PT14 PT10 PT27PT28 PT2 PT4 PT5 Graft-versus-host disease PT69 PT24 PT8 PT5 PT20PT38 PT70 PT17 PT7 PT67 PT12PT25 PT29 PT55 PT50 PT32 0510 15 20 25 30 35 PT66PT71 PT44PT43 PT62 PT22 PT37 PT58 PT33PT34 PT42 PT52PT46 PT13 PT61 −Log P PT47 PT48 PT54PT51 10 PT68PT64 PT21 PT31 PT49PT56PT45PT60 PT15 PT26 PT41 PT16 PT57PT53 PT40 EF PT18 PT65PT39

t−SNE 2 PT2 PT11 T-cell proliferation T-cell apoptotic process *** *** PT15PT23 PT69 0.75 *** *** PT1PT15PT17 PT69 0.80 PT3 PT32 *** *** PT30 PT32 PT36 PT55 PT32 PT18 PT70 PT59 PT32 *** ** PT59 PT13 PT63 PT64 PT24 PT34 PT36 PT68 PT69 PT29 PT1 PT8 PT53 PT22 PT44 PT48 PT5 PT4 PT71 PT66 PT25 PT35 PT24PT70 PT7 PT43 PT12 PT39 PT38PT47 PT65 PT23PT20 PT33PT57 PT63 PT14 PT16 PT59 PT21 PT40PT31 PT49 PT67 PT6 PT60PT45 0.70 0.75 PT6PT6 PT51PT54 PT58 PT61 PT52 PT58 PT19 PT46 PT48 PT27PT27 PT19 PT37 PT57 PT65 PT62 PT63 PT28 PT9 PT50 PT42 PT67 PT9 PT68 PT56 PT41 PT33 PT28 PT10PT11 PT44 PT36PT53 PT11 PT42 PT35 PT66 PT39 PT26 PT46PT40 PT41 PT64 PT56 PT55PT47 PT43 PT37 PT34 PT71 PT62 PT49 PT45 PT31 0.70 PT1 PT51PT54PT52 PT20 0.65 PT5 PT7 PT50 PT4

PT17 PT13 enrichment score PT29 PT3 PT22PT21 PT60 PT12PT14 PT38 PT61 PT8 Single sample GSEA PT18 PT30 PT30 PT16PT15 PT25 PT2PT10 PT26 0.65 0.60 t−SNE 1 TN TSCM TCM TEM TEFF TN TSCM TCM TEM TEFF

Figure 1. A transcriptome atlas of premanufacture T cells among children and young adults enrolled to receive anti-CD19 CAR T-cell therapy. A, T cells from 71 patients were collected at time of clinical leukapheresis and sorted into five T-cell subsets: NT , TSCM, TCM, TEM, and TEFF. All 355 T-cell populations underwent RNA-seq. For six of these patients, paired CITE-seq and scATAC-seq was performed. B, Association between proportion of TN, TSCM, TCM, TEM, and TEFF at time of leukapheresis with long-term CAR T-cell persistence, assessed by duration of BCA. Pairwise statistical significance was assessed using the Wilcoxon rank-sum test, and multiple testing correction was performed using the Benjamini–Hochberg procedure. C, t-distributed stochastic neighbor embedding (t-SNE) plot of transcriptome data, capturing the functional continuum from naïve, memory, and effector T-cell lineages. D, Enriched pathways among differentially expressed genes from comparison of T-cell subsets (ANOVA F-test FDR < 0.05, top 500 genes). E and F, ssGSEA scores of T-cell proliferation (E) and apoptotic pathways (F) across T-cell subsets. Statistical significance for pairwise comparisons was performed with Welch t test. consisting of predicted interactions between transcription fac- reviously (13). To identify T cell–specific TF–gene interactions, tors (TF) and target genes. Benchmarking studies for TRN we repeated these steps to construct a consensus TRN for non– inference have demonstrated that no individual computational T-cell immune populations from a compendium of 43 micro- method performs optimally across data sets, and a consensus array datasets from Becht and colleagues (18), representing approach achieves superior and more robust performance 708 samples, and defined a T cell–specific TRN as the network (13). Following this principle, we applied top-performing gene- whose edges are present in the T-cell TRN but not present in regulatory inference algorithms (14–17) to our expression data the non–T-cell immune TRN from public data (see Methods). to generate base networks, from which we constructed a con- To validate this network approach, we considered a set of T-cell sensus TRN using the Borda Count principle as described TFs involved in T-cell differentiation (19) as a benchmark, and

September 2021 CANCER DISCOVERY | OF4

RESEARCH ARTICLE Chen et al.

TCF7 ABPTPRK E LEF1 RIN3 PADI4 AEBP1 CD248 TCF7 MLXIP GALMBEND5 PECAM1 CLEC11A LEF1 PATZ1 CCDC106 Key TF Target gene AEBP1 ZNF496 TBXA2R ZNF496 GIPC3 MLXIP BACH2 CACHD1 PLEKHB1 DCHS1 PATZ1 DACT1 Regulon size SATB1 REG4 TMIGD2 ZNF496 APBA2 AEBP1 MYC PRAG1 MLXIP BACH2 ZNF609 NOG PATZ1 BACH2 SATB1 ZNF101 LEF1 ZNF609 9002,500 MYC ZNF414 EFNA1 EDAR TCF7 CCR7 SATB1 FAM117B ZNF609 ZNF358 KRT73 Differential expression RGS10 MYC SNN logFC ZNF101 GTF3A PRKAR1B LRRN3 GAL3ST4 MAN1C1 ZNF414 ZNF135 CD4 ZNF711 ACTN1 ZNF358 SSTR3CHMP7 −262 TCF3 AK5 LDLRAP1CA6 GTF3A FAM200B RNF157 ZNF135 SELL NELL2 NPAS2 PIK3IP1 TMEM204 ZNF711 Normalized STAT6 TCF3 expression FAM200B 00.5 1.01.5 2.0 NPAS2 1.0 Normalized regulatory potential STAT6 0.5 MAF 0.0 CDMAF TBX21 APOBEC3G CCL4 CST7 −0.5 TBX21 CLIC1 KLRB1 CEBPD −1.0 CEBPD WEE1 PRR5L Key TF Target gene ZEB2 ZEB2 CD58 ZEB2 SLC2A8 STOM PRDM1 PRDM1 PDCD1 GPR25 MSC MSC PLEKHG3 CXCR6 MYO1F PYHIN1 Regulon size IRF5 IRF5 MCOLN2 CEBPD ASB2 NR3C1 NR3C1 TGFBR3 CREM CREM TMEM273 IRF5 MSC RBPJ RBPJ ADAM19 CTNNA1 TBX21 1,1002,500 TMF1 PRDM1 GPR68 ITGB1 TMF1 BHLHE40 RCBTB2 Differential expression BHLHE40 ANXA2 MAF GNA15 logFC RORA CPNE7 FAM129A RORA TNFRSF18 ETV7 TNFRSF4 ETV7 FOSL2 DUSP4 NETO2 TIGIT CCR6 −262 FOSL2 RORC FRMD4B AIM2 NR3C1 RORC ZNF683 CLDND1CREM CCL20 ZNF683 BCL11A LRRC32 BCL11A BATF CCR4 LIMS1 BATF E2F2 RNF19A RBPJ HNRNPLL E2F2 0123 TN TSCM TCM TEM TEFF Normalized regulatory potential

Figure 2. Transcriptional regulation of naïve and effector T-cell states. A, Top 20 predicted TFs associated with TN cells sorted by normalized regulatory potential (FDR < 0.05). B, Transcriptional regulatory network of top 10 predicted TFs in A and top 50 differentially expressed genes (DEG). C, Top 20 predicted key TFs associated with TEFF cells (FDR < 0.05). D, Transcriptional regulatory network of the top 10 predicted TFs in C and top 50 DEGs. E, Differential expression of TFs predicted to regulate naïve and effector T-cell states. Log-transformed gene expression values were normalized across the 355 T-cell samples using the z-score transformation. assessed the node degree of these TFs in the T cell–specific Chronic IFN Response Associates with Poor network and the network constructed from non–T-cell public Clinical CAR T-cell Persistence immune data. The node degree of these benchmark TFs was significantly higher in the T cell–specific network compared We designed a mixed-effects regression model to identify with the null distribution (Wilcoxon rank-sum P = 7.6e-5; differentially expressed genes within T-cell subtypes associated Supplementary Fig. S5A), but not different from the null with clinical CAR T-cell persistence (Supplementary Table S2; distribution in the non–T-cell immune network (P = 0.32), Methods). This approach allowed us to identify genes and path- demonstrating the capacity of the constructed T cell–specific ways associated with clinical CAR T-cell persistence in a manner network to capture T-cell transcriptional regulatory biology. that accounts for the potential confounding effect of T-cell We have previously described a method for identifying subtype composition (Fig. 3A). We performed pathway analysis key TFs involved in regulating transcriptional fate, which on the differentially expressed genes and found that type I IFN we have recently applied to study hematopoiesis and type signaling response was the most significantly enriched pathway I diabetes (20–22). We extended this method to our T cell– (Fig. 3B, P = 2.67e-16). The genes most differentially upregu- specific network to identify key TFs acting to maintain naïve lated among patients with poor CAR T-cell persistence included and stem cell memory state, as well as TFs driving effector the type I IFN response genes RSAD2, IRF7, MX1, ISG15, OASL, T-cell development. Among the TFs with the strongest pre- and IFIT3 (FDR < 0.05; Fig. 3A). Interestingly, these genes were dicted regulatory potential driving naïve and early memory differentially expressed between patients even among the naïve T-cell state were TCF7 and LEF1 (FDR < 0.05; Fig. 2A and and memory T-cell subtypes (Fig. 3C). B), which have been previously described as essential in early We extended our gene network inference methods to define thymocyte development (23, 24) and maintenance of T-cell Sregressed, a T cell–specific network generated from expression data memory function (19, 25). Among the TFs with the strong- regressed for the potentially confounding effect of T-cell sub- est regulatory potential for effector T-cell states were TBX21 type (see Methods). We applied our TF prioritization method (T-bet), which has been associated with both effector CD8+ to identify TFs associated with clinical CAR T-cell persistence. and CD4+ Th1 differentiation (19, 26, 27); PRDM1 (Blimp-1), Our network highlighted IRF7 as the gene with the strongest which has been associated with differentiation of CD8+ cells predicted regulatory potential discriminating between clinical and non-Tfh CD4+ cells (19, 26, 28); and ZEB2, a transcrip- CAR T-cell persistence (Fig. 3D and E). Among other top hits tional repressor that has been recently shown to cooperate were STAT4, which has also been associated with IFN response with T-bet to promote a terminal CD8+ T-cell differentiation signaling and Th1 differentiation (27, 31), and SOX4, which program (refs. 29, 30; FDR < 0.05; Fig. 2C and D). These were has been associated with T-cell differentiation (32). supported with strong differential expression of these TFs Type I IFN response genes have been shown to play an across T-cell subtype (Fig. 2E; Supplementary Fig. S5B–S5F). important role in physiologic T-cell activation (33); indeed,

OF5 | CANCER DISCOVERY September 2021 AACRJournals.org

Bulk and Single-Cell Factors Mediating CAR T Persistence RESEARCH ARTICLE

RSAD2 Response to virus ABP Defense response to symbiont adj Defense response to virus FDR < 0.05 (n = 24) 7.5 IRF7 Cellular response to type I interferon MX1 FDR < 0.25 (n = 309) Type I interferon signaling pathway Response to type I interferon TYMP NR3C2 n.s. (n = 11,251) Regulation of viral life cycle ISG15 Regulation of viral process OASL CAPN12 ZFYVE9 P -value Regulation of biological process involved in symbiotic interaction THEM4 VSIG1 5.0 EPSTI1 FAM153B 10 PABPC1 Negative regulation of viral process IFIT3 HOPX TCAF1 TJP3 RPL3 ACVR2A FAM153C Protein localization to endoplasmic reticulum PABPC3 KIF5CZNF404 T−helper cell lineage commitment

−Log Protein targeting to membrane Cotranslational protein targeting to membrane 2.5 Nuclear−transcribed mRNA catabolic process, nonsense−mediated decay T−helper 17 cell lineage commitment Translational initiation SRP−dependent cotranslational protein targeting to membrane Establishment of protein localization to endoplasmic reticulum 0.0 Protein targeting to ER −10 −505 20 10 01020 −Log (FDR) Log2 fold change 10 Short CAR T persistence Long CAR T persistence Short CAR T persistence Long CAR T persistence C IRF7 (FDR = 2.0e-4) RSAD2 (FDR = 1.0e-5) MX1 (FDR = 3.0e-4) ISG15 (FDR = 6.3e-3) 1.1e-3 3.0e-4 7.8e-4 3.0e-4 0.1 1.7e-3 1.7e-3 1.7e-3 8.6e-5 9.8e-6 1.2e-2 7.3e-3 2.3e-3 1.0e-3 9.3e-4 1.1e-2 1.0e-2 7.0e-3 4.8e-3 5.3e-2 3 3 4

2 Failure of CAR 2 2 3 T-cell persistence Failure of CAR T−cell persistence Successful CAR Successful CAR T−cell persistence T-cell persistence 1 2 transcripts per 1 1 10 1 Log million reads (TPM) 0 0 0

TN TSCM TCM TEM TEFF TN TSCM TCM TEM TEFF TN TSCM TCM TEM TEFF TN TSCM TCM TEM TEFF DE IRF7 ZFYVE9 ATF4 ISG15 PARP9 ACVR2A NR3C2 TCAF1 SLC40A1 OAS1 LAPTM4B PLSCR1 CTSO KLF10 OASL IGSF22 NR3C2 SOX4 JUP HABP4 CCDC7 DRAP1 STAT4 SYDE2 ZNF404 TYMP DBP STAT4 ATF4 FAM153A ADHFE1 CEBPG SLC7A5 LY6E DRAP1 RIC3 PLSCR1 SOX4 EPSTI1 NFIA IRF7 KIF5C ZNF792 EEF1A1 Key TF Target gene IL6R YARS STAT1 RUNX2 MX1 CAPN12 VSIG1 TCF7 HOPX TJP3 Regulon size THEM4 FLI1 C2orf15 FBXL16 ZNF677 KLF10 PABPC1 TMEM204 ZBTB14 DBP IFIT3 200 2,000 MXD4 RSAD2 Differential expression abs (logFC) DR1 CEBPG CREBL2 HSD11B1L PHF7 RPL3 285 PLEKHB1 0 123 Normalized regulatory potential

Figure 3. IFN signaling pathway genes are upregulated across T-cell subsets among patients with poor CAR T-cell persistence. A, Volcano plot showing differentially expressed genes in patients with long-term (≥ 6 months) vs. failed (<6 months) CAR T-cell persistence across T-cell subsets. Significantly differentially expressed genes associated with failed CAR T-cell persistence (FDR < 0.05) are highlighted with a red box. B, Top enriched pathways among upregulated and downregulated differentially expressed genes between patients with long-term versus failed CAR T-cell persistence. C, Box plots of example differentially expressed genes in the IFN response pathway. The overall ANOVA F-test FDR is shown next to the gene. FDRs for pairwise comparisons within each T-cell subtype are shown on top of each pair of boxplots. D, Top 20 predicted key TFs associated with the long-term versus failed persistence sorted by normalized regulatory potential (FDR < 0.05). E, Transcriptional regulatory network of the top 10 predicted TFs in D and top 50 DEGs. we found that the type I IFN response pathway was upregu- that among these early memory T-cell subsets, IRF7 was the lated in the TEM and TEFF (Supplementary Fig. S6A). We asked top-ranked TF associated with poor clinical CAR T-cell per- whether this pathway was upregulated among patients with sistence, suggesting that the IFN signaling response pathways poor CAR T-cell persistence in the naïve and early memory were enriched across T-cell phenotypes, including the naïve subsets. To assess the robustness of this pathway, we repeated and early memory subsets (Supplementary Fig. S7B–S7D). our differential expression analysis on T-cell subsets that both The TCF7 Network in Patients with Long-Term included and excluded the TEM and TEFF populations. IRF7, RSAD2, MX1, and ISG15 were broadly differentially expressed CAR T-cell Persistence Is Maintained in Effector among all T-cell subsets, and IFN genes were among the top T-cell Lineages genes associated with poor CAR T-cell persistence in compari- We next hypothesized that differential molecular pathways sons including strictly naïve and early memory T-cell subsets between patients with long and short CAR T-cell persistence (Supplementary Fig. S6B). For example, in our mixed-effects may manifest in transcriptional differences in the effector interaction model, we identified 16 genes associated with T-cell phenotypes. We focused on the TEM and TEFF subsets, clinical CAR T-cell persistence within the TN, TSCM, and TCM identifying genes (Fig. 4A) and TFs (Fig. 4B) differentially subtypes, seven of which were negatively associated with expressed between patients with short and long CAR T-cell clinical CAR T-cell persistence: IRF7, RSAD2, SLC7A5, OASL, persistence. We observed an enrichment of IFN signaling TYMP, MX1, and ISG15 (Supplementary Fig. S7A). We found response pathways (P = 2.49e-8), and the most differentially

September 2021 CANCER DISCOVERY | OF6

RESEARCH ARTICLE Chen et al.

FDR < 0.05 (n = 46) FDR < 0.05 (n = 14) ABFDR < 0.25 (n = 246) FDR < 0.25 (n = 54) C n.s. (n = 11,314) n.s. (n = 956) Response to virus 8 RSAD2 TCF7 ZNF677 Defense response to virus BACH2 ZNF607 JAML TLE2 SREBF1 Defense response to symbiont SATB1 FOS 4 SNAI3 NR3C2 Cellular response to type I interferon USP44 ZBTB20 RCAN3 FBXL16 IRF7 Type I interferon signaling pathway TJP3 6 MX1 PCED1B IL23A STAT1 ZNF563 Response to type I interferon VSIG1 PLSCR1 TOB1 LY 9 THEM4 ANKRD6 Immune effector process PABPC1 FAM102ALAPTM4B TCF7 3 Interspecies interaction between organisms ITPKB CDR2 ADTRP CCR7 IL6RZNF677 Defense response to other organism IFIT3 BEX2 ZNF607 P -value MPP7SLC40A1 ACVR2A P -value Response to other organism LRP5 STMN1 MCUB CD7 10 4 ARL4A NR3C2 10 Lymphocyte activation NMT2 BACH2 MICU3 CD4 MPZL2 2 Mononuclear cell differentiation SREBF1 FOS −Log HIPK2 NMNAT3 −Log Lymphocyte differentiation Leukocyte differentiation 2 Hematopoietic or lymphoid organ development 1 Positive regulation of cell−cell adhesion Immune system development Regulation of cell−cell adhesion T cell differentiation 0 0 T cell activation −2.5 0.0 2.5 5.0 −4 −202 10 50510 Log2 fold change Log2 fold change −Log10 (FDR) Short CAR T persistence Long CAR T persistence Short CAR T persistence Long CAR T persistence Short CAR T persistence Long CAR T persistence DE ANKRD6 MX1 IFIT3 SNAI3 TCF7 ZNF607 NR3C2 MICU3 STAT1 TJP3 THEM4 ZBTB4 LY9 PCED1B BACH2 CD7 DBP DBP MPZL2 ZNF677 IL6R SATB1 RSAD2 FAM102A Key TF Target gene NMNAT3 FOS TLE2 SNAI3 LAPTM4B TTC9 STAT1 BEX2 JAML VSIG1 NMT2 ETS1 TCF7 SLC40A1 Regulon size ZBTB4 FBXL16 GATA3 IL23A PRXL2B PPARD CCR7 BACH2 ITPKB ZSCAN18 NR3C2 FOS THAP11 SATB1 TOB1 LRP5 MPP7 STMN1 ETS1 2001,200 ZBTB25 ARL4A ZNF677 HIPK2 MCUB Differential expression abs (logFC) PLSCR1 RCAN3 PABPC1 SREBF1 ADTRP FOSB TUBB2A CD4 TFDP2 ACVR2A 153 GFI1

0123 Normalized regulatory potential

Figure 4. Maintenance of the TCF7 network among effector T cells associates with long CAR T-cell persistence. A, Volcano plot displaying differentially expressed genes in TEM and TEFF between patients with long-term (≥ 6 months) versus failed (<6 months) CAR T-cell persistence. B, Volcano plot displaying differentially expressed transcription factors in TEM and TEFF between patients with long-term (≥6 months) versus failed (< 6 months) CAR T-cell persistence. C, Top enriched pathways among upregulated and downregulated differentially expressed genes between patients with long-term versus failed CAR T-cell persistence in TEM and TEFF. D, Top 20 predicted TFs associated with long CAR T-cell persistence in TEM and TEFF (FDR < 0.05). E, Transcriptional regulatory network of the top 10 predicted TFs in D and top 50 DEGs.

expressed genes associated with poor CAR T-cell persistence network analysis using exclusively TEFF expression data, and were IFN response genes including RSAD2, MX1, and IFIT3 found that TCF7 remained among the top 20 ranked TFs, (Fig. 4A). Unlike our previous analyses where IFN response along with GATA3 and BACH2 (Supplementary Fig. S8E and pathways dominated the differential expression analysis, the S8F). highest-ranked pathways for TEM and TEFF were those associated with T-cell activation and differentiation (P = 3.34e-13; The TCF7 Regulon and IFN Response Gene Fig. 4C), suggesting that regulators of T-cell differentiation Signatures Are Associated with Clinical CAR T-cell state may have an additional role in maintaining T-cell per- Therapy Outcomes in an Independent Validation Set sistence within these differentiated subtypes. We next sought to validate transcriptional gene signatures Narrowing our differential expression analysis to TFs representing the TCF7 regulon and IFN signaling response in these effector phenotypes, we found that TCF7 was the (Fig. 5A). During network construction from our complete most significantly upregulated TF associated with long CAR transcriptome dataset involving all T-cell subtypes, TCF7 was T-cell persistence (FDR = 0.018, Fig. 4B). Network analysis the transcription factor with the greatest node connectivity, revealed that TCF7 was among the top-ranked TFs associ- with 2,496 predicted targets, of which 1,487 had a positive ated with CAR T-cell persistence in TEM and TEFF; among the expression correlation with TCF7. We defined the TCF7 regu- other top hits expressed at higher levels among patients with lon as the set of predicted target genes of TCF7 with a positive long CAR T-cell persistence were BACH2, FOS, and GATA3, expression correlation, and defined the TCF7 regulon score as which is associated with Th2 development; among top hits the single-sample GSEA enrichment score using the gene set expressed at higher levels in patients with short CAR T-cell composed of TCF7 and its regulon (Supplementary Table S3). persistence was STAT1, which has been associated with Th1 Because this score was defined in a manner agnostic to T-cell development and IFN signaling (refs. 27, 34; Fig. 4D and E). type or clinical outcome labels, we first assessed whether this

Strictly within the TEFF subset, we found TCF7 was the most score discriminated between naïve and TEFF cell types, as well significantly upregulated TF associated with long CAR T-cell as between clinical response groups among TEFF cells. Indeed, persistence, and T-cell activation remained the top enriched this TCF7 regulon score, discriminated between T-cell sub- pathway (Supplementary Fig. S8A–S8D). We repeated our types, was significantly higher in EFFT among patients with

OF7 | CANCER DISCOVERY September 2021 AACRJournals.org

Bulk and Single-Cell Factors Mediating CAR T Persistence RESEARCH ARTICLE

A BCD T-cell subtype-specific RNA-seq Independent validation set: Fraietta et al. P = 6.2e-3 P = 1.7e-2 Clinical 1.0 Bulk RNA-seq response

2 2 0.8 Pre-infusion CAR T-cells: Non CAR-stimulated ( blue) 0.6 CAR-stimulated ( red) 0 0 0.4 Sensitivity

TCF7 regulon score TCF7 regulon score 0.2 ? −2 −2 Non CAR-stimulated (0.80) 0.0 CAR-stimulated (0.77) TCF7 regulon TCF7 0.80.60.40.20.0 1.0 regulon 1-specificity NR/PR CR/PRTD NR/PR CR/PRTD TCF7 score

TCF7 regulon gene signature: single-sample GSEA EFG

4 P = 3.5e-2 P = 0.92 1.0

Target genes 0.8 Genes upregulated among patients 2 Interferon with poor CART-cell persistence 2 0.6 response score 0.4 7.5 Sensitivity Interferon response 0 0 5.0 IFN response score IFN response score 0.2 P -value gene signature: 10 single-sample GSEA Non CAR-stimulated (0.71) 2.5 0.0 CAR-stimulated (0.54) −Log 1.00.80.60.40.20.0 0.0 −2 −2 −10 −5 05 1-specificity NR/PR CR/PRTD NR/PR CR/PRTD Log2 fold change

Figure 5. Validation of TCF7 regulon and IFN response gene signature in an independent data set. A, Workflow for development and validation of gene signatures representing the TCF7 regulon and IFN response pathways acting between and within T-cell subsets. B and C, Evaluation of the TCF7 regulon score on the independent dataset of Fraietta and colleagues on both the unstimulated (blue) and CAR-stimulated (red) CAR T cells from patients with chronic lymphocytic leukemia (CLL). Each point represents a patient. Clinical response groups were as previously described in the independent validation set (5), with unfavorable outcomes. NR, nonresponder; PR, partial responder; and favorable outcomes PRTD, partial responder with highly active T-cell products. CR, complete remission. Statistical significance between clinical response groups was assessed with Welcht test. D, ROC curve for the TCF7 regulon score. E and F, Evaluation of the IFN response score on the independent dataset of Fraietta and colleagues on both the unstimulated (blue) and CAR-stimulated (red) CAR T cells. Statistical significance was assessed with Welcht test. G, ROC curve for the IFN response score. long-term CAR T-cell persistence, and discriminated between Next, we developed a prognostic gene signature representing these clinical outcome groups using leave-one-patient-out the IFN response observed across T-cell subtypes. We identified and leave-one-T-cell-type-out cross-validation (Supplementary 55 genes significantly upregulated among those patients with Fig. S9A–S9D). Next, we sought to assess this score in an inde- short CAR T-cell persistence, which was strongly enriched for pendent validation set. We assessed this score on the RNA-seq type I IFN response pathways (Fig. 3A and C). We thus defined dataset of Fraietta and colleagues, which consisted of preinfu- an IFN response score as the single-sample GSEA enrichment sion anti-CD19 CAR T cells generated from adults with CLL score based on these differentially expressed genes (Supple- (5) and is, to our knowledge, the largest clinically annotated mentary Table S3). This gene signature was associated with

RNA-seq dataset from patients receiving CAR T-cell therapy T-cell subtypes and was upregulated among the TCM, TEM, and prior to this study. The CAR T cells of these 34 patients TEFF subsets compared with TN (Supplementary Fig. S9E). assessed either underwent mock stimulation (non–CAR-stim- We performed cross-validation based on a leave-one-patient- ulated group) or bead-based anti-CD19 CAR stimulation in out and leave-one-T-cell-type-out approach, finding that this vitro (CAR-stimulated group). This data set had several dif- method for gene signature discovery robustly discriminated ferences from ours, because it was generated from engineered between patients with long-term and short-term CAR T-cell CAR T cells, collected from adults with CLL, and was not persistence across T-cell subtypes [area under ROC Curve sorted by T-cell subtype. Despite these differences, we sought (AUROC) = 0.63–0.76; Supplementary Fig. S9F and S9G]. We to evaluate whether our gene signatures were robust enough next assessed this gene signature on the independent valida- to validate in this independent data set. Indeed, we found that tion set by Fraietta and colleagues (5). We found that for the our TCF7 regulon score was positively associated with favora- non–CAR-stimulated group, our IFN response signature was ble clinical response in the non–CAR-stimulated samples significantly associated with poor clinical response P( = 0.035, (P = 0.0062; Fig. 5B–D), as well as in the CAR-stimulated AUROC = 0.71; Fig. 5E). However, for the CAR-stimulated group (P = 0.017). The discriminative ability of this gene sig- group, this gene signature was not significantly prognostic P( = nature even in the CAR-stimulated group suggests that the 0.92, AUROC = 0.54; Fig. 5F and G), suggesting that the T-cell prognostic relevance of TCF7 regulon is maintained through- activation pathways triggered by CAR stimulation may negate out CAR T-cell manufacture and stimulation. the prognostic significance of the IFN response signature.

September 2021 CANCER DISCOVERY | OF8

RESEARCH ARTICLE Chen et al.

Analysis of T-cell Subset Markers and integration (Supplementary Fig. S10B and S10C). The single- TF Expression in the Manufactured cell RNA-seq (scRNA-seq) data were initially clustered into 21 CAR T-cell Product clusters based on global transcriptomic profiles, and clusters We asked to what extent the engineered CAR T cells differ were subsequently merged on the basis of selected RNA and pro- from premanufacture T cells and retain an association with tein markers (Fig. 6C). This led to the identification of 11 final + + long-term clinical CAR T-cell persistence. We identified engi- T-cell subsets, including CD4 and CD8 naïve, memory, and neered CAR T cells from 11 patients on trial to receive CAR effector subsets, as well as resting and activated regulatory T cells T-cell therapy (study identifier: NCT01626495) who were not (Treg cells) highly expressing the transcription factor FOXP3. The previously studied in our bulk RNA-seq analysis: four had long- proportion of naïve and early memory T cells ranged from 92.6% term BCA greater than or equal to 6 months, five had short-term in patient 51 with greater than 6 months of CAR T-cell persis- BCA less than 6 months, and two were not ultimately infused tence, to 72.7% in patient 66 with failed CAR T-cell persistence at due to medical contraindications. We identified premanufac- 3 months (Supplementary Fig. S10D and S10E). + ture T cells from three of these patients. FACS analysis revealed Intriguingly, we identified a cluster of CD4 T cells express- increased protein expression of IRF7 in the CAR T-cell products ing naïve and memory markers such as CCR7, CD62L, and compared with premanufacture T cells across T-cell subsets CD45RA, but also strongly expressing IFN response genes such (FDR = 0.001–0.062; Supplementary Fig. S9H), suggestive of as IRF7, RSAD2, MX1, and ISG15 as well as STAT1, which has IFN signaling induced during the CAR T-cell manufacturing been shown to be involved in IFN signaling and Th1 develop- process. Among the postmanufacture CAR T-cell samples from ment (refs. 27, 34; Fig. 6A and D; Supplementary Fig. S10F the 9 patients who received a CAR T-cell infusion, propor- and S10G). This cluster did not express other Th1 TFs, such tions of T-cell subsets defined by CD62L and CD45RO were as STAT4 or TBX21, and did not express characteristic Th2 or not significantly associated with long-term CAR T-cell persis- Treg TFs, such as GATA3, STAT6, or FOXP3 (Supplementary tence (FDR = 0.948; Supplementary Fig. S9I). RNA expression Fig. S10G). This cluster was relatively rare (approximately 1% assessed by qRT-PCR of TFs was assessed for association with of total T cells), and not significantly associated with CAR clinical CAR T-cell persistence, with expression of TCF7 trend- T-cell persistence in the n = 6 patients assessed (P = 0.40). ing toward higher expression in patients with long CAR T-cell However, we reasoned that this population may be helpful persistence, IRF7 and TBX21 trending toward higher expression in understanding the relationship between the TCF7 regulon in patients with short CAR T-cell persistence, and no clear trend and IFN response pathway. We performed AUCell analysis for LEF1 and PRDM1 expression; these trends were not statisti- (36) to define single-cell enrichment scores for the previously cally significant P( = 0.161-0.794; Supplementary Fig. S9J and defined TCF7 regulon and IFN response gene signatures. We S9K). The relatively small sample size likely contributed to the found that the TCF7 regulon score was enriched among naïve lack of statistical significance in these trends; power analysis and memory T cells, whereas the IFN response gene signature suggested that 24, 22, and 17 patients per group would be was strongly enriched among activated T cells, particularly + required to achieve statistical significance at 0.05 with 80% CD8 TEM and TEFF (Fig. 6E). These gene signatures shared a power for expression of TCF7, IRF7, and TBX21, respectively. In strong inverse relationship among the T-cell subtypes, with the + addition, in vitro stimulation during CAR T-cell manufacture notable exception of the IFN-responsive CD4 naïve/memory may at least partially abrogate between-patient T-cell differ- population, which was strongly enriched for both signatures ences, as reported in a previous study (1). (Fig. 6F). These data provide cellular level support for our prior finding that IFN response plays a role not only in activated T-cell states, but in naïve and memory states. Furthermore, Integrative Single-Cell Analysis Demonstrates this finding suggests that the TCF7 regulon and IFN response That the TCF7 Regulon is Not Mutually Exclusive pathway are not mutually exclusive, and may impart independ- to IFN Response ent effects of T-cell function and CAR T-cell efficacy. To more deeply understand the relationship between the TCF7 regulon and IFN response in premanufacture T cells, Epigenetic Regulation of TCF7 and Its Downstream we performed CITE-seq (35) and scATAC-seq on 6 of the 71 Targets Remains Partially Active in TEFF Cells patients included in this study. These 6 patients were selected We investigated the epigenetic regulation of T-cell states on the basis of sample availability and representing a range and TCF7 function through integration of scATAC-seq data of clinical CAR T-cell persistence outcomes, from failure at with CITE-seq data. We integrated patient scATAC-seq data 2 months to persistence greater than 22 months. From our using Seurat and projected these data onto the UMAP defined CITE-seq data, we obtained a dual readout of cell surface by the CITE-seq analysis. We assessed chromatin accessibility protein expression and RNA quantification within the same of TF motifs with chromVAR (37), finding that accessibility cell; scATAC-seq was performed in a separate aliquot of T cells of the TCF7/LEF1 motif was strongly associated with naïve from the same patients. After filtering out low-quality cells, our and memory T-cell states, whereas accessibility of the PRDM1 single-cell data consisted of 17,750 and 19,673 cells from the (Blimp-1) motif was associated with effector memory and CITE-seq and scATAC-seq data, respectively. We first focused effector T cells, and accessibility of the TBX21 motif was + on the CITE-seq data, performing data integration, dimension- strongly associated with CD8 TEM and TEFF (Fig. 7A and B). ality reduction, and clustering to reveal a single-cell landscape The strongest differentially accessible motif was the AP-1 of CD4+ and CD8+ T cells (see Methods; Fig. 6A–C; Supple- motif, which was closed among naïve T cells and strongly mentary Fig. S10A). The data broadly mixed by patient and opened among the TEM and TEFF subsets (Fig. 7A; Supplemen- separated by both RNA and protein markers, suggestive of good tary Fig. S11A).

OF9 | CANCER DISCOVERY September 2021 AACRJournals.org

Bulk and Single-Cell Factors Mediating CAR T Persistence RESEARCH ARTICLE

ABCD4 CD8 CCR7 C CD4+ T Activated Treg CM 2.0 2.5 + 1.5 2.0 2 CD4 T 1.5 Resting Treg EM 1.0 1.0 1 0.5 0.5 0

IFN-responsive T /T UMAP 2 N CM CD4 + CD8A + CD8 TEFF UMAP 1 CD4 TN CD8B CD62L CD45RO CD45RA CD8B2 CCR7 CD95 1.6 1.5 CD27 1.2 1.0 1.0 0.8 IL7R 0.5 0.5

UMAP 2 0.4 FOXP3 + 0.0 CD8 TCM CD4 + CD8 CD8 TSCM + CCR7 CD8 TEM CD45RO CD95 FOXP3 CD27 CD45RA CD62L CD95 3 3 N N M

1.2 EM EM T T CM 2 CM Reg 2 Reg EFF + + SCM T T T T

0.8 /TC T + + + + T N 1 1 +

0.4 + T + CD8 CD4 0.0 0 0 + CD4 CD8 CD4 CD8 T CD8 CD8

N CD8 Resting T CD4 Activated T IFN responsive

UMAP 1

DEIRF7 RSAD2 F TCF7 regulon genes + **** **** 4 **** **** CD4 T **** **** N **** **** + **** **** **** 3 **** + IFN-responsive CD4 T /T **** **** N M **** **** CD8 T 3 **** **** N **** **** 2 2 Restingst Treg 0.26 + 1 1 0.24 CD4 TCM Expression level Expression level 0.22 0 0 0.24 0.20 + MX1 ISG15 CD8 TSCM **** **** **** **** **** **** **** **** **** **** **** 6 **** + **** **** 4 **** **** CD4 T **** **** EM **** **** 4 Interferon response genes 2 + 2 0.22 CD8 TCM Expression level Expression level

0 0 AUCell score of TCF7 regulon Activated Tregeg + N N N N reg reg reg reg EM EM EM EM CD8 T CM CM CM CM CM CM T T T T EFF EFF EM + + + + SCM SCM T T T T T T T T /T /T T T 0.06 + + + + + + + + T T N N + + + + T T ated T ated ated T ated CD8 CD8 CD4 CD4 0.04 + + CD8 CD4 CD8 CD4 CD8 CD8 CD4 CD4 CD8 CD8 Resting T Resting Resting T Resting CD8 CD8 0.02 Act iv Act iv CD4 CD4 ve ve 0.20 + CD8 TEFF IFN−responsi IFN−responsi 0.02 0.04 0.06 0.08 0.10 AUCell score of IFN response gene signature

Figure 6. CITE-seq captures the heterogeneity of premanufacture T cells and demonstrates that the TCF7 regulon and IFN response gene signatures are not mutually exclusive. A, UMAP projection of 17,750 CITE-seq cells, colored by 11 T-cell clusters. B, Protein and RNA expression of selected marker genes based on CITE-seq antibody-derived tags (protein, green) and normalized RNA expression (blue). C, Cluster dendrogram of marker gene expression based on RNA expression (top) and protein expression (bottom). Complete-linkage hierarchical clustering was performed on these selected marker genes. D, Violin plots indicating the normalized RNA expression of IFN response genes for each T-cell cluster. Pairwise statistical significance was assessed using the Wilcoxon rank-sum test. ****, P < 0.0001; ***, P < 0.001; **, P < 0.01; *, 0.01, n.s., P ≥ 0.05. E, AUCell enrichment scores associated with the TCF7 regulon and IFN response gene signatures defined based on our bulk RNA-seq analysis.F, Association between TCF7 regulon and IFN response gene signatures across 11 T-cell clusters, demonstrating that these pathways are inversely related in normal T-cell differentiation (gray arrow) but not necessarily mutually exclusive, as the IFN-responsive CD4+ naïve/memory cluster was concurrently enriched in both signature scores.

While TCF7 expression and motif accessibility were dimin- Bonferroni-adjusted P = 4.3e-3), indicating that the epigenetic ished among TEFF cells (Fig. 7B), we asked whether its enhancer regulation of TCF7 remains partially active in this differenti- accessibility and downstream targets remained partially active, ated T-cell subset (Fig. 7C). Physical interactions between pre- which could provide an explanation for the positive associa- dicted enhancer–promoter interactions were validated using tion of TCF7 with clinical CAR T-cell persistence observed in chromosome conformation capture in healthy donor T cells, our bulk RNA-seq analysis. We predicted enhancer–promoter demonstrating that these predicted enhancers were strongest interactions using two methods: chromatin coaccessibil- among the naïve and early memory T-cell types and partially ity from scATAC-seq alone using Cicero, and correlation of maintained across T-cell subtypes (Supplementary Fig. S11B chromatin accessibility with RNA expression using a meta- and S11C; Supplementary Table S4). cell-based regression approach as described previously (38). Finally, we asked whether our data support the notion that Using stringent cutoffs of Cicero coaccessibility scores (>0.25) TCF7 partially maintains its regulatory interaction with its and metacell-based Bonferroni-adjusted P values (<0.05), we downstream targets in TEFF cells. We considered the set of identified three upstream enhancers with strong evidence of chromatin peaks containing the TCF7 motif, and the genes interaction with the TCF7 promoter (Fig. 7C). We repeated within the TCF7 regulon defined by our bulk RNA-seq analysis. + this analysis with the single-cell data from CD8 TEFF alone, We first considered all T cells in our scATAC-seq data, finding for which gene expression as well as chromatin accessibility that the chromatin coaccessibility scores associated with these of TCF7 and its enhancers was diminished. Despite the use TCF7 peak–promoter pairs were enriched compared with the of this restricted subset of cells, we found that one of the null distribution of non–TCF7-related promoter–peak pairs enhancers retained both chromatin coaccessibility and asso- (Supplementary Fig. S11D and S11E; Wilcoxon P = 9.01e-22). ciation with TCF7 expression (coaccessibility score = 0.37, We repeated this analysis restricted to the scATAC-seq data

September 2021 CANCER DISCOVERY | OF10

RESEARCH ARTICLE Chen et al.

ABCTCF TCF7L2 TCF7 (RNA) LEF1 (RNA) PRDM1 (RNA) TBX21 (RNA) LEF1 TCF7 BACH1 JUN::JUNB NFE2 3 3 2.5 BACH2 3 2.0 2 2 2 1.5 JDP2 1.0 FOSL1::JUND 1 1 1 0.5 JUNB 0 0 0 0.0 JUND

FOSL2::JUND UMAP 2 FOSL2::JUN FOSB::JUNB FOSL1::JUNB UMAP 1 FOSL2::JUNB FOSL1::JUN TCF7 (motif accessibility) LEF1 (motif accessibility) PRDM1 (motif accessibility) TBX21 (motif accessibility) FOS::JUN FOS::JUND FOS::JUNB FOSL1 2 2 2 BATF3 1 1 1 2 BATF::JUN 0 0 0 0 BATF −1 −1 −1 −2 JUN(var.2) −2 −2 −2 FOS

FOSL2 UMAP 2 TBX21 EOMES 2 TBX2 UMAP 1 TBR1 TBX1 1 MGA TBX15 0 C TBX20 TBX6 chr5: 134,100,000 134, 150, 000 TBX3 −1 25 _ TBX18 TBX4 CD8+ T TBX5 N RORC RORA 0 RORA(var.2) 25 _ RORB + MAF::NFE2 CD8 TSCM NFE2L1 MAFK 0 IRF9 25 _ IRF8 IRF4 + ERF CD8 TCM ETS1 FLI1 0 ERG 25 _ ELK4 + ETV5 CD8 T NFKB2 EM REL RELA 0 IRF5 25 _ HOXD4 + HOXA4 CD8 TEFF HOXC4 STAT1::STAT2 0 IRF7 IRF2 IRF3 RUNX3 EP interactions EHF ELF1 across all T-cells ELF3 ETV1 GABPA EP interactions NR2C2(var.2) SPIB within TEFF IKZF1 ETV4 TCF7 N N ive reg reg SKP1 + T + T CM CM EM EM EFF M + T SCM+ T + T + T + /T + T T CD4 CD8 N + T CD4 CD8 CD4 CD8 Resting T CD8 CD8 Activated T IFN−responsCD4

Figure 7. Single-cell ATAC-seq captures the epigenetic regulation of T-cell functional states and reveals that transcriptional regulation of TCF7 is + partially maintained in CD8 TEFF. A, Heat map of chromVAR deviation scores associated with TF motifs across T-cell clusters. Shown are motifs that were significantly associated with T-cell cluster (ANOVA FDR< 0.05) and for which at least one T-cell subtype had an absolute deviation score greater than 0.90. B, Normalized RNA expression of TCF7, LEF1, PRDM1, and TBX21, and associated chromVAR motif deviation scores. C, Chromatin accessibility + tracks for CD8 TN, TSCM, TCM, TEM, and TEFF at the TCF7 locus. The bottom tracks show high-confidence enhancer–promoter (EP) interactions predicted based on chromatin coaccessibility (Cicero score > 0.25, light red), metacell-based regression (Bonferroni-adjusted P < 0.05, light blue), and EP interactions that were concordantly predicted with both methods (dark purple). EP interactions were predicted across all T cells using the complete scATAC-seq + and CITE-seq datasets, as well as on only the CD8 TEFF cells.

+ associated with the CD8 TEFF data, asking whether the enrich- and confounded by T-cell subtype composition. By perform- ment of TCF7 regulatory interactions is maintained. While the ing subtype-specific transcriptomic analysis of 71 pediatric overall mean coaccessibility score was diminished within this patients and integrative single-cell analysis of six patients, we restricted T-cell subset compared with the pan–T-cell analysis, we developed, to our knowledge, the largest and most compre- found that the coaccessibility scores for TCF7 peak-to-promoter hensive molecular portrait of the landscape of autologously pairs remained enriched compared with the null peak-to-pro- derived T cells in CAR T-cell therapy. Our approach allowed + moter scores within the CD8 TEFF subset alone (Supplementary us to deeply characterize the composition and transcriptional Fig. S11E; P = 2.73e-3). Furthermore, the mean expression of regulation of favorable T-cell states, and, importantly, to + TCF7 targets based on peak-to-promoter pairs within CD8 TEFF account for the confounding effect of T-cell subtype to iden- was significantly greater than the background of non-TCF7 tar- tify TCF7 and IFN signaling as pathways associated with CAR get genes (P = 3.07e-22; Supplementary Fig. S11F). These data T-cell persistence. support the notion that the regulatory interactions between We found that higher proportions of naïve and early TCF7 and its targets are partially maintained even among the memory T cells were positively associated with long-term + CD8 effector T cells, providing a mechanistic explanation for clinical CAR T-cell persistence. While TEM and TEFF subsets our observation of TCF7 expression as a positive association were highly enriched in proliferative and metabolically active with clinical CAR T-cell persistence in TEFF cells. pathways—phenotypes that may superficially appear to be beneficial for effective CAR T-cell function—we observed con- comitant enrichment in intrinsic and extrinsic apoptotic DISCUSSION signaling in these subsets, suggesting that programmed cell To date, studies of the T-cell determinants of CAR T-cell death ultimately hinders the contribution of these T cells persistence have been limited by modest clinical sample sizes to long-term CAR T-cell persistence. Regulatory network

OF11 | CANCER DISCOVERY September 2021 AACRJournals.org

Bulk and Single-Cell Factors Mediating CAR T Persistence RESEARCH ARTICLE analysis nominated TCF7 and LEF1 (in contrast to other TCF potential of CAR T cells (47). This suggests that during the family members) as core TFs maintaining naïve and memory acute phase of CAR-stimulated T-cell activation, IRF7 and the T-cell states, and PRDM1 and TBX21 are associated with type I IFN network may be necessary for optimal activation

TEFF cell states; the epigenetic basis of these lineage factors and stimulation. was strongly supported by motif enrichment analysis in our Understanding the T-cell determinants of successful CAR scATAC-seq analysis. Given the evidence of a favorable role of T-cell therapy is fundamental in improving clinical outcomes naïve and early memory T cells in other CAR constructs (39) for pediatric B-cell malignancies, as well as in the develop- and in intrinsic T-cell expansion potential (4), these transcrip- ment of adoptive T-cell therapies for other malignancies tional regulators are likely of general importance to other including solid tumors. By analyzing autologously derived, forms of CAR T-cell therapy. premanufacture T cells, our unique data set provides valu- The TCF7 gene, encoding for the Tcf1 TF, has been shown able insights for generalizable T-cell mechanisms that are to profoundly remodel the T-cell chromatin landscape, and not confounded by specific perturbations in the CAR T-cell knockout studies have provided evidence that it acts as a key manufacturing process. Moreover, our analysis of sorted initiating factor in early thymocyte development (23, 40). Our T-cell subsets and single-cell analyses enabled us to identify scRNA-seq and chromatin accessibility suggest a profound gene regulatory mechanisms that account for the potential shift in the transcriptional regulatory machinery, from TCF7 confounding effect of T-cell subtype composition. Together, and LEF1 to PRDM1 and TBX21, with TBX21 expression and our data provide an expanded view of the molecular mecha- motif accessibility particularly notable among CD8+ cells. nisms associated with the long-term efficacy of clinical CAR Whether one or two TFs represent pioneering TFs that drive T-cell therapy. the rest, or if these cell fate transitions represent a cooperative TF circuit, remains unclear (19). Intriguingly, our data revealed several lines of evidence suggesting that TCF7 plays METHODS a role not only in maintaining the naïve and early memory Patient Identification and Clinical Annotation T-cell states, but also in maintaining a favorable phenotype Patients were identified by the clinical practices at the Division of in effector lineages. While TCF7 has been known to be highly Oncology, Children’s Hospital of Philadelphia (Philadelphia, PA). expressed in naïve and memory T-cell subsets, recent mouse Patients were enrolled onto Children’s Hospital of Philadelphia Insti- models of chronic viral infection and solid tumors have impli- tutional Review Board (IRB)–approved clinical trials NCT01626495 cated TCF7 in cell fate decisions in effector and exhausted and NCT02906371, with written informed consent obtained by phenotypes. For example, recent experimental studies have patients or their guardians in accordance with the U.S. Common led to the discovery of PD1+Tcf1+Tim3− progenitor exhausted Rule. Premanufacture T cells were acquired from the IRB-approved T cells; TCF7 appears to be a critical regulator of maintain- local institutional trial CHP-784. For event-free survival analyses, we defined a CAR T-cell failure ing stemness in exhausted T cells, and may be responsible event as the clinical progression of leukemic blast cells or detection of for the proliferative burst seen in anti–PD-1 therapy (41–43). circulating B cells. Event-free survival was recorded as a censored data Our findings represent a key translation to a novel clinical point if the patient had BCA and proceeded to receive a bone marrow domain, suggesting that TCF7 acts within TEFF cells to main- transplant, died of other causes such as cytokine release syndrome, had tain a favorable T-cell phenotype not only in chronic viral CD19− relapse, or had continued BCA at the most recent follow-up. infection and solid tumors, but in other forms of immunotherapy including adoptive T-cell immunotherapy. T-cell Enrichment and Cell Sorting The IFN response pathway has been shown to play a major T cells were enriched from apheresis samples by negative selection role in T-cell function and dysfunction, and its role depends using EasySep Human T Cell Enrichment Kit (#19051; StemCell strongly on the context and duration of response. Although Technologies) according to the manufacturer’s instructions. Previ- IFN signaling stimulates T-cell responses in the short term, ous work (48, 49) has demonstrated that CCR7, CD62L, CD45RO, chronic IFN signaling has been shown in experimental mod- and CD95 can be used to differentiate the various T-cell phenotypes + + els to drive T-cell dysfunction (44). IRF7, a key regulator of the by using the following expression patterns: TN – CCR7 , CD62L , − − + + − + type I IFN response (45), emerged in our data as a transcrip- CD45RO , CD95 ; TSCM – CCR7 , CD62L , CD45RO , CD95 ; + + + + − − tional regulator associated with poor CAR T-cell persistence TCM – CCR7 , CD62L , CD45RO , CD95 ; TEM – CCR7 , CD62L , + + − − − + across T-cell subsets. Our data suggest an immunosuppres- CD45RO , CD95 ; TEFF – CCR7 , CD62L , CD45RO , CD95 . Briefly, cells were resuspended in FACS buffer (Ca++ and Mg++ free PBS + 1% sive effect of chronic IFN signaling among premanufacture BSA), then incubated with CCR7 antibody for 20 minutes at 37°C, T cells that contributes to failure of long-term CAR T-cell washed once, and then incubated with remaining antibody cocktails persistence. IFN signaling was associated with poor CAR for 25 minutes at 4°C. Samples were then washed twice, and sorting T-cell persistence even in naïve and early memory T cells, was done using MoFlo Astrios EQ High-Speed Cell Sorter (Beckman suggesting that this pathway is likely driven not by repeated Coulter, Inc). The antibodies used for cell sorting described above cognate antigen binding, as in T-cell exhaustion, but by alter- were CCR7-FITC (#561271; BD Biosciences), CD95-PE (#556641; BD native mechanisms such as circulating inflammatory factors. Biosciences), CD45RO-BV421 (#304224; BioLegend), and CD62L- Mouse models of persistent LCMV infection have shown PerCP/Cyanine5.5 (#304824; BioLegend). that blockage of type I IFN is associated with improved T-cell responses (33, 46), highlighting the immunosuppres- Bulk RNA-seq sive effect of chronic IFN signaling. In contrast, Zhao and Cells were sorted into TRIzol-LS (Invitrogen), and total RNA was colleagues found that CAR stimulation strongly induces IRF7 purified using RNeasy Micro Kit (cat no. 74004, Qiagen) and analyzed expression, and that knockdown of IRF7 reduced the cytolytic for purity and integrity using RNA 6000 Pico Kit for Bioanalyzer

September 2021 CANCER DISCOVERY | OF12

RESEARCH ARTICLE Chen et al.

2100 (catalog no. 5067–1513, Agilent). Full-length cDNA was syn- Authors’ Contributions thesized and amplified using SMART-seq V4 Ultra Low Input RNA G.M. Chen: Formal analysis, validation, investigation, visualiza- Kit (catalog no. 634891, Takara) from 1 ng total RNA per sample. tion, methodology, writing–original draft, writing–review and editing. Full-length cDNA was purified using SPRIselect beads (catalog no. C. Chen: Resources, data curation, investigation, writing–review and B23318, Beckman Coulter). Sequencing libraries were prepared using editing. R.K. Das: Data curation, investigation, writing–review and Nextera XT DNA Library Prep Kit (catalog no. FC-131–1096) and 200 editing. P. Gao: Data curation, investigation, writing–review and edit- pg purified full-length cDNA per sample. Libraries were sequenced ing. C. Chen: Data curation, investigation, writing–review and editing. on an Illumina Hiseq 2500 in paired-end mode with the read length S. Bandyopadhyay: Formal analysis, validation. Y. Ding: Writing– of 100 nt. review and editing. Y. Uzun: Software, methodology. W. Yu: Software, methodology. Q. Zhu: Software, methodology. R.M. Myers: CITE-seq Investigation. S.A. Grupp: Conceptualization, resources, writing– Sorted cells were blocked with Human TruStain FcX (BioLegend, review and editing. D.M. Barrett: Conceptualization, resources, catalog no. 422301) and then stained with a TotalSeq-A antibody supervision, funding acquisition, investigation, writing–original panel (see Supplementary Data). Stained cells were immediately draft, project administration, writing–review and editing. K. Tan: processed using the 10x Genomics Chromium controller and the Conceptualization, resources, supervision, funding acquisition, Chromium Single Cell 3′ reagent kits (V3). 3′ GEX libraries were investigation, methodology, writing–original draft, project adminis- constructed using 10x Genomics library preparation kit. ADT librar- tration, writing–review and editing. ies were constructed using KAPA HiFi HotStart ReadyMix kit (Kapa Biosystems, catalog no. KK2601). Library quality was checked using Acknowledgments Agilent High Sensitivity DNA kit and Bioanalyzer 2100. Libraries We thank the Children’s Hospital of Philadelphia Flow Cytometry were quantified using dsDNA High-Sensitivity (HS) assay kit (Invit- Core for their assistance with cell sorting, and the Research Informa- rogen) on Qubit fluorometer and the qPCR-based KAPA quantification Services for providing computing support. This work was sup- tion kit. Libraries were sequenced on an Illumina Nova-Seq 6000 with ported by NIH grants CA232361 (to D.M. Barrett and S.A. Grupp) 28:8:0:87 paired-end format. and CA233285 (to K. Tan), a Doris Duke Charitable Foundation Clinical Scientist Development Award and a Stand Up To Cancer scATAC-seq Innovative Research Grant, Grant Number SU2C-AACR-IRG 12-17 Sorted cells were centrifuged at 300 × g for 5 minutes at 4°C. (to D.M. Barrett), a CIHR Doctoral Foreign Study Award #433117 Forty-five microliters of chilled lysis buffer was added to cell pellets (to G.M. Chen), a NIH/National Child Health and Human Devel- and mixed by pipetting gently three times, and incubated 3 minutes opment award T32HD043021, and NCI award K12CA076931 (to on ice. After incubation, 50 μL of prechilled wash buffer was added Y. Ding), NIH Medical Scientist Training Program T32 GM07170 without mixing and centrifuged immediately at 300 × g for 5 minutes (to S. Bandyopadhyay), and a 2020 Blavatnik Family Fellowship in at 4°C. Ninety-five microliters of supernatant was carefully discarded Biomedical Research (to Q. Zhu). Stand Up To Cancer (SU2C) is a and 45 μL prechilled diluted nuclei buffer (10x Genomics) was added division of the Entertainment Industry Foundation. The indicated without mixing and sample was centrifuged at 300 × g for 5 minutes SU2C research grant is administered by the American Association for at 4°C. The nuclei pellet was then resuspended in 7 μL prechilled Cancer Research, a scientific partner of SU2C. diluted nuclei buffer, and nuclei concentration was determined using a Countess II cell counter (Invitrogen). Nuclei (7,000–20,000) Received November 17, 2020; revised March 3, 2021; accepted April were used for the transposition reaction in bulk, and then loaded 1, 2021; published first April 5, 2021. to the 10x Genomics Chromium controller and processed with the Chromium Single Cell ATAC Reagent Kit. Library quality was checked using Agilent High Sensitivity DNA Kit and Bioanalyzer References 2100. Libraries were quantified using dsDNA High-Sensitivity (HS) . 1 Finney OC, Brakke HM, Rawlings-Rhea S, Hicks R, Doolittle D, assay kit (Invitrogen) on Qubit fluorometer and the qPCR-based Lopez M, et al. CD19 CAR T cell product and disease attributes pre- KAPA Quantification Kit. Libraries were sequenced on an Illumina dict leukemia remission durability. J Clin Invest 2019;129:2123–32. Nova-Seq 6000 with 49:8:16:49 paired-end format. 2. Neelapu SS, Locke FL, Bartlett NL, Lekakis LJ, Miklos DB, Jacobson Bulk and single-cell data have been submitted to dbGaP under CA, et al. Axicabtagene ciloleucel CAR T-cell therapy in refractory Study Accession phs002323.v1.p1. large B-cell lymphoma. N Engl J Med 2017;377:2531–44. 3. Cohen AD, Garfall AL, Stadtmauer EA, Melenhorst JJ, Lacey SF, Additional methods are described in the Supplementary Methods. Lancaster E, et al. B cell maturation antigen–specific CAR T cells are clinically active in multiple myeloma. J Clin Invest 2019;129: Authors’ Disclosures 2210–21. S.A. Grupp reports grants, personal fees, and other support from 4. Singh N, Perazzelli J, Grupp SA, Barrett DM. Early memory pheno- Novartis; grants from Kite and Servier, grants and other support types drive T cell proliferation in patients with pediatric malignan- from Vertex; personal fees from Roche, GSK, Humanigen, CBMG, cies. Sci Transl Med 2016;8:320ra3. 5. Fraietta JA, Lacey SF, Orlando EJ, Pruteanu-Malinici I, Gohil M, and Janssen/JnJ; and other support from Jazz, Adaptimmune, TCR2, Lundh S, et al. Determinants of response and resistance to CD19 chi- Cellectis, Juno, Allogene, and Cabaletta outside the submitted work; meric antigen receptor (CAR) T cell therapy of chronic lymphocytic in addition, S.A. Grupp has a patent for Toxicity management for leukemia. Nat Med 2018;24:563–71. antitumor activity of CARs, WO 2014011984 A1 issued. D.M. Barrett 6. Xu Y, Zhang M, Ramos CA, Durett A, Liu E, Dakhova O, et al. Closely reports he began work while employed at the Children’s Hospital of related T-memory stem cells correlate with in vivo expansion of CAR. Philadelphia. Between the time of submission and acceptance, D.M. CD19-T cells and are preserved by IL-7 and IL-15. Blood 2014;123: Barrett left CHOP and is now an employee of Tmunity Therapeutics, 3750–9. Inc. D.M. Barrett reports that his employment had no relation to the 7. Maude SL, Frey N, Shaw PA, Aplenc R, Barrett DM, Bunin NJ, et al. submitted work and no financial conflicts are present. No disclosures Chimeric antigen receptor T cells for sustained remissions in leuke- were reported by the other authors. mia. N Engl J Med 2014;371:1507–17.

OF13 | CANCER DISCOVERY September 2021 AACRJournals.org

Bulk and Single-Cell Factors Mediating CAR T Persistence RESEARCH ARTICLE

8. Maude SL, Laetsch TW, Buechner J, Rives S, Boyer M, Bittencourt H, 29. Omilusik KD, Best JA, Yu B, Goossens S, Weidemann A, Nguyen JV, et al. Tisagenlecleucel in children and young adults with B-cell et al. Transcriptional repressor ZEB2 promotes terminal differentia- lymphoblastic leukemia. N Engl J Med 2018;378:439–48. tion of CD8+ effector and memory T cell populations during infec- 9. Newman AM, Steen CB, Liu CL, Gentles AJ, Chaudhuri AA, Scherer F, tion. J Exp Med 2015;212:2027–39. et al. Determining cell type abundance and expression from bulk tis- 30. Dominguez CX, Amezquita RA, Guan T, Marshall HD, Joshi NS, sues with digital cytometry. Nat Biotechnol 2019;37:773–82. Kleinstein SH, et al. The transcription factors ZEB2 and T-bet cooper- 10. Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W, et al. limma ate to program cytotoxic T cell terminal differentiation in response to powers differential expression analyses for RNA-sequencing and LCMV viral infection. J Exp Med 2015;212:2041–56. microarray studies. Nucleic Acids Res 2015;43:e47. 31. Nguyen KB, Watford WT, Salomon R, Hofmann SR, Pien GC, 11. Law CW, Chen Y, Shi W, Smyth GK. voom: precision weights unlock Morinobu A, et al. Critical role for STAT4 activation by type 1 inter- linear model analysis tools for RNA-seq read counts. Genome Biol ferons in the interferon-gamma response to viral infection. Science 2014;15:R29. 2002;297:2063–6. 12. Barbie DA, Tamayo P, Boehm JS, Kim SY, Moody SE, Dunn IF, et al. 32. Kuwahara M, Yamashita M, Shinoda K, Tofukuji S, Onodera A, Systematic RNA interference reveals that oncogenic KRAS-driven Shinnakasu R, et al. The transcription factor Sox4 is a downstream cancers require TBK1. Nature 2009;462:108–12. target of signaling by the cytokine TGF-β and suppresses T(H)2 dif- 13. Marbach D, Costello JC, Küffner R, Vega NM, Prill RJ, Camacho DM, ferentiation. Nat Immunol 2012;13:778–86. et al. Wisdom of crowds for robust gene network inference. Nat Meth- 33. Teijaro JR, Ng C, Lee AM, Sullivan BM, Sheehan KCF, Welch M, et al. ods 2012;9:796–804. Persistent LCMV infection is controlled by blockade of type I inter- 14. Faith JJ, Hayete B, Thaden JT, Mogno I, Wierzbowski J, Cottarel G, feron signaling. Science 2013;340:207–11. et al. Large-scale mapping and validation of Escherichia coli tran- 34. Gough DJ, Messina NL, Hii L, Gould JA, Sabapathy K, Robertson APS, scriptional regulation from a compendium of expression profiles. et al. Functional crosstalk between type I and II interferon through PLoS Biol 2007;5:e8. the regulated expression of STAT1. PLoS Biol 2010;8:e1000361. 15. Huynh-Thu VA, Irrthum A, Wehenkel L, Geurts P. Inferring regula- 35. Stoeckius M, Hafemeister C, Stephenson W, Houck-Loomis B, Chat- tory networks from expression data using tree-based methods. PLoS topadhyay PK, Swerdlow H, et al. Simultaneous epitope and tran- One 2010;5:e12776. scriptome measurement in single cells. Nat Methods 2017;14:865–8. 16. Haury A-C, Mordelet F, Vera-Licona P, Vert J-P. TIGRESS: trustful 36. Aibar S, González-Blas CB, Moerman T, Huynh-Thu VA, Imrichova H, inference of gene REgulation using stability selection. BMC Syst Biol Hulselmans G, et al. SCENIC: single-cell regulatory network inference 2012;6:145. and clustering. Nat Methods 2017;14:1083–6. 17. Bonneau R, Reiss DJ, Shannon P, Facciotti M, Hood L, Baliga NS, 37. Schep AN, Wu B, Buenrostro JD, Greenleaf WJ. chromVAR: infer- et al. The Inferelator: an algorithm for learning parsimonious regula- ring transcription-factor-associated accessibility from single-cell epi- tory networks from systems-biology data sets de novo. Genome Biol genomic data. Nat Methods 2017;14:975–8. 2006;7:R36. 38. Zhu Q, Gao P, Tober J, Bennett L, Chen C, Uzun Y, et al. Developmen- 18. Becht E, Giraldo NA, Lacroix L, Buttard B, Elarouci N, Petitprez F, tal trajectory of prehematopoietic stem cell formation from endothe- et al. Estimating the population abundance of tissue-infiltrating immune lium. Blood 2020;136:845–56. and stromal cell populations using gene expression. Genome Biol 2016; 39. Sommermeyer D, Hudecek M, Kosasih PL, Gogishvili T, Maloney DG, 17:218. Turtle CJ, et al. Chimeric antigen receptor-modified T cells derived 19. Chang JT, Wherry EJ, Goldrath AW. Molecular regulation of from defined CD8+ and CD4+ subsets confer superior antitumor effector and memory T cell differentiation. Nat Immunol 2014;15: reactivity in vivo. Leukemia 2016;30:492–500. 1104–15. 40. Johnson JL, Georgakilas G, Petrovic J, Kurachi M, Cai S, Harly C, 20. Gao L, Tober J, Gao P, Chen C, Tan K, Speck NA. RUNX1 and the et al. Lineage-determining transcription factor TCF-1 initiates the endothelial origin of blood. Exp Hematol 2018;68:2–9. epigenetic identity of T cells. Immunity 2018;48:243–57. 21. Gao P, Uzun Y, He B, Salamati SE, Coffey JKM, Tsalikian E, et al. 41. Im SJ, Hashimoto M, Gerner MY, Lee J, Kissick HT, Burger MC, et al. Risk variants disrupting enhancers of TH1 and TREG cells in type 1 Defining CD8+ T cells that provide the proliferative burst after PD-1 diabetes. Proc Natl Acad Sci U S A 2019;116:7581–90. therapy. Nature 2016;537:417–21. 22. Gao P, Chen C, Howell ED, Li Y, Tober J, Uzun Y, et al. Transcriptional 42. Wu T, Ji Y, Moseman EA, Xu HC, Manglani M, Kirby M, et al. The regulatory network controlling the ontogeny of hematopoietic stem TCF1-Bcl6 axis counteracts type I interferon to repress exhaustion cells. Genes Dev 2020;34:13–4. and maintain T cell stemness. Sci Immunol 2016;1:eaai8593. 23. Yu S, Zhou X, Steinke FC, Liu C, Chen S-C, Zagorodna O, et al. 43. Miller BC, Sen DR, Al Abosy R, Bi K, Virkud YV, LaFleur MW, et al. The TCF-1 and LEF-1 transcription factors have cooperative and Subsets of exhausted CD8+ T cells differentially mediate tumor control opposing roles in T cell development and malignancy. Immunity and respond to checkpoint blockade. Nat Immunol 2019;20:326–36. 2012;37:813–26. 44. Minn AJ, Wherry EJ. Combination cancer therapies with immune check- 24. Weber BN, Chi AW-S, Chavez A, Yashiro-Ohtani Y, Yang Q, Shestova O, point blockade: convergence on interferon signaling. Cell 2016;165:272–5. et al. A critical role for TCF-1 in T-lineage specification and differen- 45. Ning S, Pagano JS, Barber GN. IRF7: activation, regulation, modifica- tiation. Nature 2011;476:63–8. tion and function. Genes Immun 2011;12:399–414. 25. Zhou X, Xue H-H. Cutting edge: generation of memory precursors 46. Wilson EB, Yamada DH, Elsaesser H, Herskovitz J, Deng J, Cheng G, and functional memory CD8+ T cells depends on T cell factor-1 and et al. Blockade of chronic type I interferon signaling to control persis- lymphoid enhancer-binding factor-1. J Immunol 2012;189:2722–6. tent LCMV infection. Science 2013;340:202–7. 26. Best JA, Blair DA, Knell J, Yang E, Mayya V, Doedens A, et al. Tran- 47. Zhao Z, Condomines M, van der Stegen SJC, Perna F, Kloss CC, scriptional insights into the CD8(+) T cell response to infection and Gunset G, et al. Structural design of engineered costimulation deter- memory T cell formation. Nat Immunol 2013;14:404–12. mines tumor rejection kinetics and persistence of CAR T cells. Cancer 27. Oestreich KJ, Weinmann AS. Master regulators or lineage-specifying? Cell 2015;28:415–28. Changing views on CD4+ T cell transcription factors. Nat Rev Immu- 48. Gattinoni L, Lugli E, Ji Y, Pos Z, Paulos CM, Quigley MF, et al. A nol 2012;12:799–804. human memory T cell subset with stem cell-like properties. Nat Med 28. Lu KT, Kanno Y, Cannons JL, Handon R, Bible P, Elkahloun AG, et al. 2011;17:1290–7. Functional and epigenetic studies reveal multistep differentiation 49. Das RK, Vernau L, Grupp SA, Barrett DM. Naïve T-cell deficits at and plasticity of in vitro-generated and in vivo-derived follicular T diagnosis and after chemotherapy impair cell therapy potential in helper cells. Immunity 2011;35:622–32. pediatric cancers. Cancer Discov 2019;9:492–9.

September 2021 CANCER DISCOVERY | OF14

Integrative Bulk and Single-Cell Profiling of Premanufacture T-cell Populations Reveals Factors Mediating Long-Term Persistence of CAR T-cell Therapy

Gregory M. Chen, Changya Chen, Rajat K. Das, et al.

Cancer Discov Published OnlineFirst April 5, 2021.

Updated version Access the most recent version of this article at: doi:10.1158/2159-8290.CD-20-1677

Supplementary Access the most recent supplemental material at: Material http://cancerdiscovery.aacrjournals.org/content/suppl/2021/04/19/2159-8290.CD-20-1677.DC2 http://cancerdiscovery.aacrjournals.org/content/suppl/2021/04/06/2159-8290.CD-20-1677.DC1

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://cancerdiscovery.aacrjournals.org/content/early/2021/08/14/2159-8290.CD-20-1677. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.