RNA polymerase II pausing factor NELF in CD8+ T cells promotes antitumor immunity

T cell factor 1 (TCF1) is required for memory and stem-like CD8+ T cell functions. How TCF1 partners with other transcription factors to regulate transcription remains unclear. Here we show that negative elongation factor (NELF), an RNA polymerase II (Pol II) pausing factor, cooperates with TCF1 in T cell responses to cancer. Deletion of mouse Nelfb, which encodes the NELFB subunit, in mature T lymphocytes impairs immune responses to both primary tumor challenge and tumor antigen-mediated vaccination. Nelfb deletion causes more exhausted and reduced memory T cell populations, whereas its ectopic expression boosts antitumor immunity and efficacy of chimeric antigen receptor T-cell immunotherapy. Mechanistically, NELF is associated with TCF1 and recruited preferentially to the enhancers and promoters of TCF1 target genes. Nelfb ablation reduces Pol II pausing and chromatin accessibility at these TCF1-associated loci. Our findings thus suggest an important and rate-limiting function of NELF in anti-tumor immunity.

T cells undergo rapid proliferation and acquire effector function after encountering antigens, co-stimulation signals, and inflammatory cytokines 1-3 . Short-lived effector cells often undergo apoptosis after the clearance of foreign antigens. A small subset of effector cells develops into long-lived memory T cells 2 . Both effector and memory T cells play critical roles in controlling virus infection and tumor outgrowth. However, unlike acute virus infection, tumor antigens often persist in an excessive amount and thus induce an exhaustion phenotype of tumor-reactive T cells 4 . Exhausted T cells, which exhibit upregulation of multiple inhibitory receptors and loss of polyfunctionality, have been the major target for reinvigorating T cell function in anticancer immunotherapy 3 . Despite clinical success in adoptive cell therapy and immune checkpoint blockade-based therapeutics 5,6 , most patients still cannot benefit from these immunotherapies, partly due to the lack of persistent tumorreactive memory T cells.
Recent studies reported a number of key transcription factors that control memory T cell differentiation, including TCF1 7 , FOXO1 8 , MYB 9 , BACH2 10 , and BATF3 11 . In particular, TCF1 protein, encoded by the gene TCF7, plays a critical role in the regulation of T cell development 12 , differentiation 13 , and effector function preservation during exhaustion 14,15 . TCF1 promotes chromatin accessibility that favors memory T cell differentiation 16 . Meanwhile, TCF1 depletion exacerbates, and enforced TCF1 expression ameliorates T cell exhaustion 15,16 . TCF1 also promotes long-term T cell survival by promoting the anti-apoptotic factor BCL2 15 while suppressing the pro-apoptotic factor BIM 16 . A TCF1 + T cell population is critical in response to immune checkpoint blockade immunotherapy 17 . Although the role of TCF1 in the determination of T cell fate is well established, little is known about the partners that facilitate its function in T cell differentiation-related transcriptional programming.
Pol II pausing plays a pivotal role in regulating metazoan gene expression [18][19][20] . At the molecular level, promoter-proximal Pol II pausing prevents nucleosome assembly and thus maintains a permissive chromatin structure for future rounds of transcriptional activation in response to environmental stimuli 21 . Recent studies also showed that Pol II pausing and release play a role in transcriptional enhancers 22 . NELFB is one of the four subunits of the NELF complex that controls Pol II pausing 23 . Mouse genetic studies demonstrate that NELFB is indispensable for early embryogenesis 24 and mammary gland development 25 . In addition, NELF is involved in various aspects of adult tissue homeostasis including energy metabolism in the myocardium 26 , inflammatory responses in macrophages 27 , myofiber repair after injury 28 , and maintenance of junctional integrity 29 . However, the functional significance of NELF-dependent Pol II pausing in the context of T cell function remains unclear.
Here, we define a CD8 + T cell-intrinsic role for NELF during antitumor immune response. Using tissue-specific mouse genetic models, we demonstrate that NELFB in CD8 + T lymphocytes is important for antitumor immunity. Mechanistically, NELF is associated with TCF1 and facilitates chromatin accessibility at TCF1-bound transcriptional enhancers and promoters. We further establish that ectopic NELFB expression boosts host antitumor immunity in mouse models and the efficacy of CAR-T immunotherapy.

Results
Nelfb is required for antitumor CD8 + T cell function. To investigate the role of NELF in mature T cell function, we deleted Nelfb in a mature T cell-specific manner by crossing Nelfb f/f and distal Lck-Cre (dLck-Cre) mouse strains. Unlike the proximal Lck-Cre system, dLck-Cre is only activated after thymocyte positive selection and therefore has minimal impact on thymocyte development 30,31 . NELFB protein levels were significantly reduced in primary CD8 + T cells, but not in non-CD8 + cells, from Nelfb f/f ,dLck-Cre knockout (hereafter KO) mice versus their Nelfb f/f counterparts (Supplementary Fig. 1a, b). Levels of the other three NELF subunits were also decreased in KO mice versus the Nelfb f/f control ( Supplementary  Fig. 1b), consistent with the previous finding that protein stability of the four NELF subunits is mutually dependent 32 . KO mice had a slightly lower percentage of CD8 + and increased CD4 + T cells than Nelfb f/f controls ( Supplementary Fig. 1c). Cell numbers of CD4 + subpopulations, including naïve, central memory, effector memory, and regulatory T (Treg) cells, were comparable between Nelfb f/f and KO ( Supplementary Fig. 1d, e). Among CD8 + T cells, the number of naïve T cells was comparable between Nelfb f/f and KO mice (Supplementary Fig. 1f). However, central memory and effector memory T cell populations were significantly reduced in KO mice compared to those in Nelfb f/f mice ( Supplementary Fig. 1f), which could be indicative of a defective response of these mice to environmental antigens.
To investigate the role of NELF in T cell response to tumor growth, we challenged Nelfb f/f and KO mice with syngeneic mouse mammary tumor cells (E0771 and AT-3). Tumors grew more robustly in KO versus Nelfb f/f hosts ( Fig. 1a-d). Tumorinfiltrating lymphocytes (TILs) contained substantially fewer total leukocytes (Fig. 1e) and CD8 + T cells (Fig. 1f) of KO versus Nelfb f/f tumor-bearing mice. In addition, KO mice had smaller effector memory (Fig. 1g) and proliferative CD8 + T cell populations (Fig. 1h). To ascertain intrinsic defects of CD8 + T cells from KO mice, we adoptively transferred CD8 + T cells from Nelfb f/f or KO mice to immunodeficient Rag1 −/− recipient mice and subsequently challenged the chimeric mice with E0771 tumors. As expected, mice that received Nelfb f/f CD8 + T cells exhibited a strong antitumor response compared to those that received vehicles (Fig. 1i, j). In contrast, adoptively transferred KO CD8 + T cells did not confer any tumor-inhibiting activity versus controls (Fig. 1i, j). Tumors harvested from chimeric mice with KO CD8 + T cells had significantly smaller total and effector memory CD8 + populations versus those with Nelfb f/f CD8 + T cells (Fig. 1k, l). These findings demonstrate a CD8 + T cellintrinsic function of NELF in antitumor immunity.
Nelfb deletion impairs memory T cell recall response. Memory T cell response is critical for mediating the protective role of vaccination 33,34 . To determine the impact of Nelfb KO on memory function during tumor antigen-initiated vaccination, we chose two aggressive, ovalbumin (OVA)-expressing tumor models: lymphoma E.G7-OVA and melanoma B16-OVA ( Fig. 2a and Supplementary Fig. 2a). In both tumor models, nonvaccinated (non-vac), tumor-bearing KO mice tended to have worse survival than their Nelfb f/f counterparts, though the difference was not statistically significant ( Fig. 2b and Supplementary Fig. 2b). This is likely because the robust tumor growth outstripped host antitumor immunity. When mice were vaccinated with OVA protein and then challenged with tumor cells, vaccination significantly lengthened survival in both Nelfb f/f and KO hosts. However, vaccination exhibited substantially greater protection against tumors in Nelfb f/f than KO mice (p = 0.002 in Fig. 2b, p = 0.01 in Supplementary Fig. 2b). In addition, vaccination markedly increased CD8 + T cell abundance in Nelfb f/f , but not KO, hosts (Fig. 2c), suggesting that Nelfb KO impairs memory T cell response.
To ascertain the role of NELF in memory T cell function, we next used heat-inactivated B16 tumor cells as the vaccine based on an established protocol 35  in non-vaccinated Nelfb f/f and KO mice (Fig. 2d). In contrast, vaccination provided robust survival benefits in Nelfb f/f , but not in KO, mice (Fig. 2d), further corroborating the importance of NELF in memory T cell response. Following vaccination, tumorbearing Nelfb f/f hosts, but not KO ones, displayed significant increases in CD8 + T cell abundance in the spleen, and central memory T cells in both spleen and lymph nodes (Supplementary Fig. 2c-e). Taken together, our studies from multiple models of tumor antigen-initiated vaccination demonstrate that NELF plays a pivotal role in promoting tumor antigen-initiated T cell recall response.
Nelfb KO causes functional defects in CD8 + T cells. To further characterize CD8 + T cell defects in Nelfb KO mice, we sought to define cellular functions affected by Nelfb loss during T cell receptor (TCR) activation. Carboxyfluorescein succinimidyl ester (CFSE) labeling showed that KO CD8 + T cells had relatively normal proliferation rates shortly after in vitro TCR activation (days 1-4), but exhibited significant proliferation defects on day 5 ( Fig. 3a and Supplementary Fig. 3a). This indicates that Nelfb deletion unlikely directly impaired TCR-activated cell cycle entry. On the other hand, significantly more KO CD8 + T cells underwent apoptosis compared to their Nelfb f/f counterparts as early as 24 hr after TCR activation ( Fig. 3b and Supplementary Fig. 3b). Interleukin 2 (IL2) is critical for the expansion of activated T cells but also promotes an exhaustion phenotype 36 . As expected, under prolonged in vitro incubation with IL2, both Nelfb f/f and KO CD8 + T cells gradually increased expression of the immune inhibitory receptors PD1 and TIM3 (Fig. 3c-f). Notably, the magnitude of increase in either PD1/TIM3 single or doublepositive exhausted populations was significantly higher in KO versus Nelfb f/f cells (Fig. 3c-f). Loss of polyfunctionality is a hallmark of reduced memory stem cell population and increased T cell exhaustion 37 . Extended in vitro proliferation also resulted in a substantially lower abundance of interferon γ (IFNγ) and tumor necrosis factor α (TNFα) double-positive, polyfunctional KO T cells versus Nelfb f/f (Fig. 3g, h and Supplementary Fig. 3c, d). In aggregate, our data strongly indicate that Nelfb ablation exacerbates CD8 + T cell exhaustion during ex vivo expansion.
To confirm the impact of Nelfb deletion on CD8 + T cell functionality, we performed single-cell RNA-sequencing (scRNAseq) using total CD8 + T cells isolated from Nelfb f/f and KO mice. Based on canonical marker gene expression patterns, we identified four major subsets-naïve, memory, exhausted, and senescent populations ( Fig. 4a and Supplementary Fig. 4a).
Compared to Nelfb f/f CD8 + T cells, KO samples had smaller naïve and memory, but larger exhausted and senescent, cell populations ( Fig. 4b and Supplementary Fig. 4b-d). Using an established algorithm for trajectory analysis 38 , we found two differentiation trajectories initiated from the naïve stage; one leads to the exhausted stage and the other to the senescent stage (Fig. 4c). Pseudotime is a parameter that describes the positioning of individual cells from the differentiation starting point along the specific trajectory, therefore representing the degree of the differentiation from the naïve to terminally differentiated stage 39 . Pseudotime analysis of the differentiation trajectory clearly showed that for both exhausted and senescent lineages, KO cells had significantly higher pseudotimes versus Nelfb f/f (Fig. 4d). This indicates that KO cells are further downstream from their differentiation origin and therefore are more terminally differentiated than Nelfb f/f cells. Together, our data support the notion that NELF prevents precocious terminal differentiation of CD8 + T cells.
Preferential NELF-dependent Pol II pausing at TCF1 targets. To elucidate the molecular mechanism by which NELF regulates T cell functionality, we performed NELFB ChIP-seq in primary mouse CD8 + T cells. An unbiased analysis of transcription factor binding motif enrichment showed that NELFB chromatin binding overlapped most significantly with that of TCF1 (Fig. 5a) 40   , KO non-vac (n = 6), KO vac (n = 9); one-way-ANOVA for comparing mean differences. Tumor curves were compared using two-way ANOVA followed by multiple comparisons. Source data are provided as a Source Data file. targets 13,41 versus non-TCF1 targets (Fig. 5b). Next, we conducted Pol II ChIP-seq in primary Nelfb f/f and KO CD8 + T cells. We used Pol II pausing index to assess the degree of promoterproximal enrichment of Pol II, and value D to denote the maximum difference in vertical distance between two cumulative distributions 42,43 (Fig. 5c). In Nelfb f/f CD8 + T cells, TCF1 targets had a significantly higher Pol II pausing index versus non-TCF1 targets (Fig. 5c, D = 0.45736, p < 2.2e-16). Nelfb KO decreased the Pol II pausing index to a greater extent at TCF1 targets (D = 0.3172, p < 2.2e-16) than at non-TCF1 targets (D = 0.13387, p < 2.2e-16, Fig. 5c, Supplementary Fig. 5a). When enhancers and TSS were analyzed separately, the reduction in Pol II binding upon Nelfb KO was more pronounced at TCF1 targets than at TCF1 non-bound targets ( Fig. 5d and Supplementary Fig. 5b). Together, our data indicate that NELF-dependent Pol II binding in primary CD8 + T cells is preferentially associated with TCF1bound regulatory regions.
NELF regulates chromatin accessibility of TCF1 targets. Pol II accumulation at promoters and enhancers have been implicated in chromatin accessibility 21,44 . We, therefore, used an assay for transposase-accessible chromatin sequencing (ATAC-seq) to assess global chromatin openness in primary naive Nelfb f/f and   h Percentages of Nelfb f/f and KO CD8 + T cells after 10 days of in vitro culture, n = 3/group. Data were presented as mean ± SD; mean differences were compared using Student's t-test. Time-dependent curves were compared using two-way ANOVA followed by multiple comparisons. Two-sided tests were used. Source data are provided as a Source Data file.  Fig. 5d), suggesting a role of NELF in promoting chromatin accessibility. Reminiscent of its enrichment in NELFB chromatin binding regions ( Fig. 5a), TCF1 binding was the most significantly enriched motif in the chromatin regions with KO-impaired accessibility (Fig. 6a). Nelfb KO-triggered reduction in chromatin openness occurred more at TCF1 targets than non-TCF1 targets, and furthermore, the KO effect was more pronounced at enhancers than TSS of TCF1 targets ( Fig. 6b and Supplementary Figs. 5e, 6a, b). These data support the notion that NELF in naïve CD8 + T cells preferentially facilitates chromatin accessibility at TSS and to a greater extent, enhancers of TCF1 targets. The propensity of NELF chromatin binding and its action for TCF1 targets prompted us to discern a physical relationship between these two transcription factors. TCF1 protein levels were comparable in naïve Nelfb f/f and KO CD8 + T cells (Supplementary Fig. 6c), suggesting that NELF unlikely impacts TCF1 targets by regulating its expression. Using proximity labeling 45 , we found that NELF was in close proximity with TCF1 (Fig. 6c). As a positive control, similar physical proximity was detected between NELFB and NELFE, another NELF subunit (Fig. 6c). This finding is consistent with the idea that NELF and TCF1 work at a common set of targets to regulate chromatin accessibility in naïve CD8 + T cells.
Chromatin accessibility at transcriptionally regulatory regions often precedes transcription of associated genes 46,47 . We, therefore, conducted deep RNA-sequencing using Nelfb f/f and KO CD8 + T cells before and after in vitro TCR activation by anti-CD3/CD28 plus IL2. While baseline transcriptomes were similar between Nelfb f/f and KO cells, TCR-activated transcriptomes of WT and KO cells were quite distinct (Fig. 6d, Supplementary  Fig. 6d, and Supplementary Data 1). Of note, gene set enrichment analysis (GSEA) showed that Nelfb deletion was associated with enriched gene signatures for TCF1 deficiency, T cell exhaustion, and aging signature (Fig. 6e). Furthermore, KO cells exhibited reduced gene signatures for memory T cells and fatty acid metabolism, a hallmark of memory T cells 48 (Fig. 6e). In keeping with our findings, GSEA analysis of published scRNA-seq data from human melanoma showed that high NELFB expression in tumor-infiltrating CD8 + cells significantly correlated with gene signatures for memory T cell and fatty acid metabolism 49 (Fig. 6f). Thus, our data strongly suggest that NELF facilitates TCR-triggered transcription that favors memory T cell fate and mitigates T cell exhaustion and aging.
NELFB overexpression boosts antitumor immunity. To determine whether NELFB overexpression could enhance adaptive immunity, we established a T cell-specific transgenic mouse model (referred to as Tg hereafter, Supplementary Fig. 7a). In addition to elevated NELFB levels (Fig. 7a), the expression of NELFA and NELFC, two other NELF subunits, were also increased in CD8 + T cells of Tg mice ( Supplementary Fig. 7b), likely through stabilization of the entire NELF complex. In an in vivo competitive assay, KO or Tg CD45.2 + CD8 + T cells were mixed at a 1:1 ratio with their corresponding control CD8 + T cells carrying a congenic marker CD45.1 + , which were subsequently transferred into B16 tumor-bearing recipient mice. Tumor-infiltrating CD8 + cells were analyzed 2-3 weeks posttransfer (Fig. 7b)  outcompeted by control cells (Fig. 7c, d). In contrast, Tg cells comprised most of the tumor-infiltrating CD8 + cell population (Fig. 7e, f), indicating that NELFB-overexpressing T cells are superior to their WT counterparts. When purified WT and Tg CD8 + cells were transferred separately into Rag1 −/− immunodeficient hosts followed by E0771 tumor challenge, Tg CD8 + T cells again exhibited more potent antitumor activity than their WT counterparts (Fig. 7g-i). Furthermore, mice receiving Tg CD8 + cells had significantly more total leukocytes (Fig. 7j) and CD8 + TILs (Fig. 7k). NELFB overexpression increased effector memory (Fig. 7l) and proliferative CD8 + T cell populations (Fig. 7m) while reducing the exhaustion marker PD1 expression (Fig. 7n). In addition, NELFB overexpression increased both single and double IFNγ + /TNFα + CD8 + cells (Fig. 7o-q). Collectively, our data suggest that NELF is a rate-limiting factor in boosting CD8 + T cell antitumor activity, which likely occurs through reducing T cell exhaustion and increasing memory and polyfunctionality.
T cell exhaustion and the lack of sustained persistence are the major barriers to successful CAR-T therapy 50 . We, therefore, sought to determine whether human NELFB (hNELFB) can boost T cell functionality in a more clinically relevant setting. We engineered a bi-cistronic CD19-specific CAR vector based on an established CAR construct, anti-CD19-28z 51 (anti-CD19-28z-   Fig. 8b-d). In vitro expanded human T cells with hNELFB overexpression displayed an increased percentage of CD62L + CD45RA + , a reported hallmark shared by naïve and memory stem cells 52 , in both CD4 + and CD8 + populations (Fig. 8a, b). In a Raji lymphoma model using NSG immunodeficient mice, T cells carrying the parental anti-CD19-CAR-28z vector significantly prolonged the survival of tumor-bearing mice when compared with mice receiving PBS or mock-infected T cells (Fig. 8c). Of note, T cells with anti-CD19-CAR-28z-hNELFB conferred markedly superior host survival benefits over those with the parental CAR-T vector (Fig. 8c). In a solid tumor model, in which CD19 antigen was engineered in human breast cancer cell line MDA-231, host mice receiving hNELFB-expressing CAR-T cells exhibited smaller tumor growth than those with parental CAR-T cells (Supplementary Fig. 8e). Furthermore, compared to parental CAR-T, hNELFB-expressing CAR-T conferred more tumor infiltration of CD8 + and CD4 + T cells ( Supplementary Fig. 8f, g), and higher memory marker CD127 expression and fewer cells with the exhaustion markers TIM3 + CD39 + in both CD8 + and CD4 + populations (Supplementary Fig. 8h-k). Our data, therefore, provide the proof of principle that hNELFB overexpression could boost CAR-T anticancer immunotherapy.
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-29869-2 ARTICLE tissue-specific Nelfb KO mouse models [26][27][28] . It is conceivable that both the protein amount of individual NELF subunits and the stability of the NELF complex can affect T cell lifespan and functionality. The robust phenotypes of our KO and transgenic mouse models provide compelling evidence for an important and rate-limiting role of NELF in sustaining CD8 + cell functions in antitumor immunity. APEX2-mediated proximal labeling confers cross-linking of neighboring proteins in the radius of 1-10 nm 55 . Therefore, our result strongly suggests physical proximity, but not necessarily direct protein-protein interaction, between NELFB and TCF1. Nevertheless, together with the genome-wide findings, the proximal labeling data are consistent with the notion that NELF and TCF1 are part of the same transcription complex co-occupied at a common set of transcriptional enhancers and promoters in CD8 + T cells. It is also worth pointing out that our ATAC-seq was conducted in primary naïve CD8 + T cells, whereas bulk RNA-seq was done in CD8 + T cells before and after TCR activation. Unlike baseline transcriptomes prior to TCR activation, which were similar between Nelfb f/f and KO cells, TCR-activated transcriptomes of Nelfb f/f and KO cells were substantially divergent. Therefore, our data favor the model in which NELF-dependent chromatin openness prior to TCR activation precedes and likely facilitates TCR-stimulated transcription of TCF1 target genes (Fig. 8d).
NELF-dependent Pol II pausing at promoter-proximal regions is known to facilitate transcription by maintaining an accessible chromatin architecture 21 and stabilizing the transcription initiation complex 26 . In contrast, relatively little is known about the functional consequence of Pol II accumulation at transcriptional enhancers. Our genome-wide survey of primary CD8 + T cells supports the notion that NELF-dependent Pol II accumulation has even stronger effects on chromatin accessibility in enhancers versus promoters. While the exact mechanism by which enhancer-associated Pol II regulates transcription remains to be determined, we envision at least two plausible scenarios: (1) enhancer-associated Pol II may facilitate enhancer-promoter looping/interaction (Fig. 8d); and (2) Pol II-transcribed enhancer RNA could serve as the anchor for additional transcription factors/cofactors 44,56 . Histone-based epigenetic programming has been implicated in various stages of T cell development and differentiation 57 . For example, terminally exhausted and stemlike memory T cells have distinctive histone modification and chromatin accessibility profiles 58 . In addition, histone methyltransferase Suv39h1 has been shown to restrain the transcriptional program of memory T cell differentiation by enabling H3K9me3 deposition at memory-related genes 59 . We propose that NELF-mediated Pol II pausing may be part of the epigenetic programming that dictates the differentiation and function of memory T cells.
Our work suggests a strategy to improve outcomes of CD8 + T cell-based adoptive cell therapy. NELFB overexpression in both human and mouse T cells significantly enhances antitumor immunity and prolongs the survival of tumor-bearing hosts. Consistent with our findings, TCF1 overexpression has been shown to boost antitumor immunity and thus proposed to enhance CAR-T efficacy 60 . We propose that NELF belongs to an expanding list of transcription factors whose overexpression bolsters CAR-T efficiency 52 transcription factors regulate the same or distinct transcriptional programs in T cell-based immunotherapy. The anti-CD19-28z CAR construct we used is more exhaustion-prone than newer generations of anti-CD19 CAR constructs carrying the 4-1BB or ICOS signaling domain 62 . To strengthen the translational potential of NELFB overexpression in boosting adoptive cell therapy, future experiments are needed to test NELFB overexpression in other CAR constructs with tonic signaling and exhaustion-prone phenotypes, such as GD2.28z CAR-T 63 . Because CD8 + T cell exhaustion was observed as a pan-cancer phenotype 64 , NELF function in mitigating immune exhaustion could be applied to multiple cancer types.
In summary, our current study shows that NELF in CD8 + T cells works at both enhancers and promoters of TCF1 target genes to potentiate chromatin accessibility and transcriptional activation. We propose that the functional cooperation between NELF-dependent Pol II pausing and TCF1 is a mechanism that controls conversion between memory T cells and differentiated effector cells, ultimately dictating antitumor immunity and efficacy of cell-based immunotherapy against cancer.

Methods
Mice. All animal protocols were approved by the Institutional Animal Care and Use Committee at George Washington University. Nelfb f/f and distal Lck-cre (dLck-cre) mice were created as previously described 24,65 . Nelfb Tg mice were generated by CRISPR-based gene editing (Cyagen Biosciences). Briefly, the gRNA to ROSA26, a donor vector containing CAG promoter-loxP-Stop-loxP-mouse Nelfb cDNA-polyA, and Cas9 mRNA were co-injected into fertilized mouse eggs to generate targeted knockin offspring. F0 founder animals, identified by PCR followed by sequence analysis, were bred to WT mice and assessed for germline transmission and F1 mouse generation. T cell-specific Nelfb KO or transgenic mice were generated by crossing dLck-cre with Nelfb f/f or Nelfb Tg , respectively. The Cre and floxed strain behaved the same as WT B6 mice; we, therefore, used the floxed mice as WT. Rag1 −/− (stock no. 002216), NSG (stock no. 005557), and C57BL/6 CD45.1 congenic mice (stock no. 002014) were purchased from The Jackson Laboratory. Mice that are 10-week-old were considered adults and were used for tumor studies.
Heat-inactivated tumor cell vaccination experiments were done using an established protocol with some modifications 35 . Briefly, 5 × 10 6 heat-killed (boiled for 30 min) B16 mouse melanoma cells (ATCC, CRL-6475) were subcutaneously inoculated into the back flanks of mice. Two weeks after the initial vaccination, 5 × 10 6 heat-killed B16 cells were subcutaneously inoculated into the back flanks of mice as a booster. Following another 2 weeks, 1 × 10 5 live B16 cells were subcutaneously inoculated into the back flanks of mice and tumor growth was monitored.
OVA immunization was done following a previously published protocol 67 . Briefly, OVA emulsified in CFA (Hooke Laboratories, EK-0301) was subcutaneously injected at two sites on the backs of mice (0.05 ml per site). Two weeks later, OVA emulsified in IFA (Hooke Laboratories, EK-0311) was subcutaneously injected at one site on the backs of mice (0.1 ml) as a booster. Control CFA emulsions (Hooke Laboratories, CK-0301) and IFA emulsions (Hooke Laboratories, CK-0311) were used as the corresponding controls. After at least 2 weeks of booster injection, 2 × 10 6 E.G7-OVA mouse T cell lymphoma cells (ATCC, CRL-2113) or 2.4 × 10 6 B16-OVA mouse melanoma cells (generated by Dr. Tyler Curiel's lab at the University of Texas Health San Antonio) were subcutaneously injected into the backs of mice. All solid tumors were measured by digital calipers (tumor volume = 0.5 × length × width 2 ). Survival analyses used animal death or tumor size >1000 mm 3 as endpoints.
In vitro CD8 + T cell characterization. For apoptosis detection, naïve CD8 + T cells were isolated from mouse splenocytes using a negative selection protocol (STEMCELL Technologies, 19858). Cells were then activated by anti-CD3/CD28 (25 μL beads per million cells) (Thermo Fisher, 11452D) following the manufacturer's instructions. After 24 h of activation, cells were stained using an Annexin V Apoptosis Detection Kit (eBioscience, 88-8007) and analyzed by flow cytometry.
For CFSE labeling experiments, isolated naïve CD8 + T cells were first labeled with a CFSE Cell Division Tracker Kit (BioLegend, 50-712-280) and activated by anti-CD3/ CD28 (Thermo Fisher, 11452D) for 24 h. The activation beads were then removed by a magnet and recombinant mouse IL2 (R&D SYSTEMS, 402-ML-020) was added to the culture medium (10 ng/ml). Fresh IL2-containing medium was replaced every other day. Cell seeding density was maintained between 0.5-1 × 10 6 /ml. CFSE intensity was examined by flow cytometry.
To characterize cytokine production and exhaustion phenotypes, total CD8 + T cells were first isolated from mouse splenocytes (STEMCELL Technologies, 19853). Cells were then activated by anti-CD3/CD28 (Thermo Fisher, 11452D) and subsequently cultured in an IL2-containing (R&D Systems, 402-ML-020) medium. Freshly prepared IL2-containing medium was replaced every other day. Anti-PD1 and anti-TIM3 were stained and examined by flow cytometry at the indicated timepoints. For cytokine production, cells were treated with BD GolgiPlug (BD Biosciences, 550583) at 37°C for 5 h and permeabilized using a BD Cytofix/ Cytoperm kit (BD Biosciences, 554714) and subsequently stained with anti-IFNγ and anti-TNFα.
Bulk RNA-seq and single-cell RNA-seq analyses. For bulk RNA-seq, total CD8 + T cells directly isolated from 8-month-old mouse splenocytes were studied in the pre-activation stage. CD8 + T activated by anti-CD3/CD28 and subsequently expanded in IL2-containing medium for 8 days were used as the post-activation stages. Total RNA was extracted by an RNeasy Mini Kit (Qiagen, 74104) following the manufacturer's instructions. RNA samples were processed further as follows. Briefly, about 500 ng total RNA was used for library preparation following the protocol for Illumina TruSeq stranded mRNA-seq. PolyA-containing mRNA were enriched and converted into first-strand cDNA by random primers and reverse transcriptase. Second-strand cDNA were then synthesized and final RNA-seq libraries were generated by PCR. An Illumina HiSeq 3000 platform was used to carry out 50 bp single-read sequencing.
For single-cell RNA-seq, total CD8 + T cells were isolated from 10-month-old mouse spleens using a negative selection protocol according to the manufacturer's instructions (STEMCELL Technologies, 19853). Freshly isolated, >90% viable CD8 + T cells were immediately loaded onto a Chromium Next GEM Chip G (10x Genomics, PN-1000127) and processed with the Chromium Controller (10x Genomics), with a target of 3000 cells per sample. Single-cell RNA-seq libraries were built using Chromium Next GEM Single Cell 3′ GEM, Library & Gel Bead Kit v3.1 (10x Genomics, PN-1000128), and barcoded with unique Illumina sample indexes (10x Genomics, PN-120262). Libraries were sequenced on an Illumina Hiseq instrument with a 10x Genomics-compatible configuration, and 50,000 reads per cell were targeted.
Sequencing data were demultiplexed and processed by Cell Ranger (version 3.1.0). STAR 68 was used to map reads to the mouse reference genome (refdata-cell ranger-mm10-3.0.0). The outputs of individual samples were loaded into the Seurat R package 69 (version 3.1.5). High-quality cells were filtered based on the number of genes detected (between 1000 and 5000) and the percentage of unique molecular identifiers (UMIs) mapped to mitochondrial genes (<12%). Individual samples were integrated, and principal components were calculated. The first ten principal components were used for cell clustering and tSNE visualization. Data QC analysis for scRNA-seq were included in Supplementary Fig. 9a-d. Bimod test was used for marker gene expression comparison between different populations ( Supplementary  Fig. 4c, d).
ChIP-seq and ATAC-seq assays. For ChIP-seq assays, cells were first cross-linked using 1% formaldehyde for 10 min and subsequently terminated by 125 mM glycine at room temperature for 5 min. The cross-linking reagent was removed by spinning at 1000 g at 4°C for 5 min. Cells were then washed with cold PBS three times. From this step until ChIP elution, all buffers were prepared with a freshly added cocktail of protease inhibitors (1 μg/ml leupeptin, 1 μg/ml aprotinin, 1 μg/ml pepstatin, and 1 mM phenylmethylsulfonyl fluoride). Cells were lysed on ice for 10 min using lysis buffer (5 mM HEPES, pH 7.9, 85 mM KCl, 0.5% Triton X-100). The supernatant was removed after spinning at 1600 × g at 4°C for 5 min, and pellets were resuspended with nuclei lysis buffer (50 mM Tris-HCl, pH 8.0, 10 mM EDTA, 1% SDS). Chromosomal DNA was sonicated using a probe sonicator on ice, then centrifuged at 14,000 × g for 15 min, and the supernatant saved for immunoprecipitation. 10% of the sonicated DNA was saved as input. Antibodies used for ChIP include anti-Pol II (BioLegend; 664906) and anti-NELFB (Cell Signaling Technology, 14894 S). Sonicated DNA was incubated with antibodies at 4°C overnight. Dynabeads Protein A (Thermo Fisher Scientific, 10002D) was added the following day and incubated for 2 h. After incubation, Dynabeads was washed as previously described 42 . Samples were subsequently eluted and reverse-cross-linked at 65°C overnight. Immunoprecipitated chromatin and input chromatin were ethanol-precipitated and used for library construction.
Sequencing and bioinformatic analysis were conducted at the UT Health San Antonio Genome Sequencing Facility. Briefly, the size distribution of ChIP-DNA was checked by Fragment Analyzer High Sensitivity DNA assays (Agilent Technologies). 0.1-20 ng ChIP-DNA (100-400 bps) was used for ChIP-seq library preparation using SwiftBiosciences Accel-NGS 2 S Plus DNA Library Kit (SwiftBiosciences, 21024). ChIP-seq libraries were quantified by Qubit and Bioanalyzer, and then pooled for cBot amplification and sequenced with 50 bp single-read sequencing using an Illumina HiSeq 3000 platform. After sequencing, fastq files were generated with Bcl2fastq2. ChIP-seq data quality was checked by MultiQC (v1.9).
For ATAC-seq assays, freshly isolated naïve CD8 + T cells from mouse spleen (STEMCELL Technologies, 19858) were washed and centrifuged at 500 × g for 5 min. Cell pellets were then resuspended in a cryopreservation solution containing FBS and 10% DMSO. Approximately 100,000 frozen cells of each sample were used by the Active Motif to perform ATAC-seq and analyses. Briefly, cells were thawed at 37°C. Cell nuclei were first isolated to perform Tn5 tagmentation and make libraries of open chromatin, as previously described 70,71 . Tagmented DNA was then purified using a MinElute PCR purification kit (Qiagen, 28004), amplified with ten cycles of PCR, and purified using Agencourt AMPure SPRI beads (Beckman Coulter). Afterward, paired-end (PE42) sequencing with a depth of 30 million reads (a total of 60 million reads) was performed using NextSeq 500 (Illumina). FRIP scores and peak counting were used as quality controls. Reads were aligned to the mouse genome (mm9) using the BWA algorithm. Duplicate reads were removed, and only uniquely mapped reads (mapping quality ≥1) and reads mapping as matched pairs were used for further analysis. Peaks were identified using the MACS 2.1.0 algorithm. Peak read numbers correlation were included in Supplementary Fig. 9e. For genomic analyses, active enhancers were defined by H3K27ac binding peaks at non-TSS regions in CD8 + T cells, and active promoters/TSS were defined as H3K4me3 peaks that overlap with TSS in CD8 + T cells 72 . TCF1 targets were defined using previously published CD8 + TCF1 ChIPseq data 41 .
Proximity labeling and western blotting. APEX2-mediated proximity biotinylation was done as previously described 73 . Briefly, APEX2-encoding DNA sequences were fused to the mouse Nelfb gene using standard molecular cloning techniques, and the overexpression plasmid was packaged with helper plasmids in 293 T cells to generate lentivirus stocks. Afterward, Jurkat cell lines (ATCC, TIB-152) were infected with the lentivirus and a stable cell line was selected using neomycin. Biotin-phenol labeling was conducted with 30 min incubation in 500 μM biotin-phenol. Cells were then exposed to 1 mM H 2 O 2 at room temperature for 1 min. The reactions were then stopped with ice-cold Dulbecco's phosphate-buffered saline with quenchers (10 mM sodium azide, 10 mM sodium ascorbate, and 5 mM Trolox). Cells were then pelleted by centrifugation and lysed by RIPA lysis buffer containing 1 mM PMSF, 5 mM Trolox, 10 mM sodium ascorbate, and 10 mM sodium azide. A slurry of streptavidin magnetic beads (NEB, S1420S) was incubated with cell lysate and rotated at room temperature for 1 h. The beads were subsequently washed and boiled to elute biotinylated proteins. Cells without H 2 O 2 exposure served as negative controls. The resultant protein lysates were analyzed using standard Western blotting techniques. Primary antibodies included anti-TCF1 (CST, 2203 S), anti-NELFE (Proteintech, 10705-1-AP), and anti-GAPDH (Bio-Rad, 12004167). Other antibodies used for Western blotting included anti-NELFB (Cell Signaling Technology, 14894 S), anti-NELFA (Proteintech, 10456-1-AP), and anti-NELFC (Cell Signaling Technology, 12265 S). All antibodies were used at 1:1000 dilution.
CAR-T generation and characterization. A lentiviral vector pELPS-CAR19-28z was a generous gift from Dr. Carl H. June's lab at the University of Pennsylvania 51 . The human NELFB gene coding sequence was inserted downstream of the CAR19-28z-encoding sequence, with the P2A cleavage sequence in between (Gene Universal, Inc.). Ultra-purified high-titer viruses for both unmodified and modified lentiviral vectors were packaged and generated by VectorBuilder Inc.
Statistics. Mean differences between the two groups were tested using Student's ttest. Mean differences between three or more groups were tested using one-way ANOVA. Tumor curves were compared using two-way ANOVA followed by multiple comparisons. Two-sided tests were used. Survival analyses were done by Log-rank (Mantel-Cox) and Gehan-Breslow-Wilcoxon tests. GSEA was done using GSEA software 74 . To analyze human melanoma tumor-infiltrating lymphocytes (TIL), we downloaded the single-cell RNA-seq dataset from GSE72056 49 . Activated CD8 + TILs (CD8a ≥5 and CD44 ≥2) were used for GSEA as previously described 75 . Population distributions in scRNA-seq data analyses were examined using Chi-square tests. Wilcoxon rank-sum test with continuity correction was used for genomic analysis. Statistics were performed using GraphPad Prism software. p < 0.05 was considered significant.