Programmable CAR-engaging particles rewire human CAR T-cell signaling for sustainable ex vivo expansion

Abstract

T cell proliferative capacity and persistence determine the therapeutic efficacy of chimeric antigen receptor (CAR) T cells. However, strategies to externally enhance CAR-T cell expansion without genetic rewiring are lacking. Here, we engineer CAREp, programmable DNA-scaffolded PLGA microparticles displaying CAR-targeting antigens and CD28-costimulatory antibodies, to repeatedly stimulate human CD8⁺ CAR-T cells ex vivo. CAREp sustains a-EGFR CAR-T cell expansion for over 100 days across both 4-1BBζ and CD28ζ constructs, achieving up to 10¹⁸-fold cumulative expansion and surpassing tumor-cell or CD3/CD28-Dynabeads stimulation. Expanded cells retain effector function and mitochondrial fitness while exhibiting clonal enrichment, initially preserved memory-associated progenitor states, delayed exhaustion, and transiently activated telomerase delaying telomere attrition. Early transcriptomic responses show coordinated activation of DNA repair, chromatin remodeling, telomere maintenance, and mitochondrial function while restricting differentiation-associated signaling programs, mirroring long-term functional outcomes. These findings demonstrate that nanoscale ligand organization synchronizes acute CAR-T signaling with durable proliferative and metabolic states.

T cell proliferative capacity and persistence are key determinants of CAR-T cell therapeutic efficacy. Here, the authors show that CAREp, a microparticle-based platform displaying CAR-targeting antigens and CD28-costimulatory antibodies, selectively and sustainably expands human CD8⁺ CAR-T cells while maintaining a memory-associated progenitor state and delayed exhaustion.

Introduction

Chimeric antigen receptor (CAR) T cell therapies have transformed treatment of relapsed and refractory B-cell malignancies¹. Despite their clinical success, major challenges remain, most notably, limited T cell persistence after infusion^1–4 and inconsistent proliferative capacity during ex vivo manufacturing and in vivo expansion^5–8. Understanding the conditions that govern CAR-T cell proliferation and persistence is therefore critical for overcoming these challenges. Particularly, how external signals can be structured to support long-term CAR-T cell expansion and function remains poorly defined.

Physiological T cells possess intrinsic capacity for longevity and tightly regulated proliferation^9–12. Currently, genetic engineering approaches enforce intrinsic modifications by bypassing critical cell-cycle checkpoints and thereby risk oncogenic transformation^13–18. Naturally, memory T cells transiently upregulate telomerase activity upon activation to preserve telomere length and prevent senescence during expansion^19–23. Massive and sustained proliferation has been demonstrated in natural T cells when stimulated with appropriately balanced antigen and costimulatory inputs^9,11,12. These findings raise a key question: can CAR-T cells be externally stimulated, without genetic perturbations, to access a similar proliferative and functional potential? Moreover, while CAR-T cells respond acutely to antigens, costimulatory ligands, and cytokines, chronic stimulation often results in dysfunction and exhaustion^4,24–26. It remains unclear how the composition and organization of extracellular cues affect intracellular signaling and long-term fate decisions in engineered T cells. A mechanistic framework linking early signaling dynamics to durable cell states is needed to better understand and potentially modulate CAR-T cell behavior.

We hypothesize that precisely structured extracellular signals, delivered through programmable synthetic materials−DNA-scaffolded poly(lactic-co-glycolic acid) (PLGA) microparticles^27,28, could reshape CAR-T cell signaling and long-term states. To test this hypothesis, here we create CAR-engaging particles (CAREp) composed of CAR-target antigens and CD28-costimulatory antibodies (a-CD28 Ab) presented at tunable stoichiometries. CAREp is designed to function downstream of conventional CAR engineering workflows by providing programmable CAR-specific stimulation after CAR expression has been established (Fig. 1a). These microparticles are used to repeatedly stimulate human CD8⁺ CAR-T cells in vitro, with each round of stimulation initiated only after cells have returned to a resting state from the prior stimulation (~8 days) to minimize overstimulation-associated dysfunction. As comparison conditions, CD3/28-Dynabeads are included as a benchmark for the current standard approach used in T cell manufacturing, while an EGFR⁺ glioblastoma U87 cell line is used to model CAR-T cell responses during repeat encounters with antigen-expressing tumor cells. Importantly, both comparison conditions are evaluated using the same pulsed restimulation framework, in which cells are restimulated only after returning to a resting phase (Fig. 1a).

Fig. 1. PLGA microparticles functionalized with antigens and CD28 agonist antibodies enabled durable and massive EGFR CAR-T cell expansion in vitro.

a Schematic of the CAREp platform and experimental workflow. Human T cells were engineered with conventional CAR lentiviral methods and rested prior to stimulation. DNA-scaffolded PLGA microparticles were functionalized with CAR antigens and CD28 agonists at tunable ratios to create CAR-engaging particles (CAREp). b Compared to conventional stimulation by tumor cells, CAREp rewired acute T cell signaling, upregulating pathways for proliferation, metabolism, and telomere maintenance while downregulating canonical differentiation and exhaustion signaling. These rewired early signals aligned with long-term functional outcomes, including massive expansion, memory-like progenitor enrichment, robust metabolic fitness, and transiently stabilized telomere length, demonstrating CAR-T fate and function governed by precise extracellular cues. c Cumulative expansion of human CD8⁺ EGFR 4-1BBζ CAR-T cells among three donors under repeated CAREp stimulations every 8-10 days when cells are rested from the previous stimulation. The fold of expansion refers to the yielded number of T cells relative to the number present at the time of initial CAREp stimulation. d Representative 4-1BBζ and CD28ζ CAR surface expression by Myc-tag staining on the expanded CD8⁺ T cells after repeated CAREp stimulations (from n = 3 donors). e Cumulative expansion folds of CD8⁺ 4-1BBζ-CAR (from n = 4 donors) and CD28ζ-CAR (from n = 6 donors) T cells after 3 and 6 rounds of CAREp stimulation. f Cumulative expansion of CD8⁺ 4-1BBζ-CAR T cells repeatedly stimulated by a glioblastoma cell line U87 supplemented at 1:1 cancer-to-T cell ratio versus by CAREp when cells are rested from the previous stimulation. g Cumulative expansion of 4-1BBζ and CD28ζ CAR-T cells stimulated by 6 rounds of CAREp versus CD3/28-Dynabeads when cells are rested from the previous stimulation. P values were determined by two-tailed paired t-test. Data in (f,g) are from n = 3 healthy donors.

Remarkably, we find that repeated CAREp stimulation sustains prolonged ex vivo expansion of human CAR-T cells while preserving functional competence, metabolic fitness, and memory-associated cell states. Mechanistically, CAREp rewires early signaling programs associated with DNA repair, chromatin remodeling, telomere maintenance, and mitochondrial function that parallel long-term functional outcomes. We further demonstrate that transient telomerase activation and delayed telomere attrition accompany sustained expansion, linking extracellular cue organization to intrinsic programs governing CAR-T cell longevity (Fig. 1b). These findings establish nanoscale ligand organization as a programmable regulator of CAR-T cell signaling and fate. More broadly, this work provides a framework for engineering therapeutic immune cells through external signal programming rather than synthetic genetic rewiring.

Results

Massive and sustained expansion of human CAR-T cells from CAREp-stimulations

Agonist antibodies against the CD28 receptor can provide essential costimulatory signaling for T cell survival and proliferation^29,30. To test if the coating of EGFR (epidermal growth factor receptor) antigens on PLGA particles can stimulate CAR-T cell proliferation and if the co-presentation of a-CD28 Ab can augment the activation, we engineered micron-sized CAREp particles with different ratios of the two signals (at 1:0, 9:1, 3:1, 1:3, and 0:1) on the surfaces, leveraging the robust and precise control of ligand functionalization through synthetic 17-base DNA scaffolds (Fig. 1a and Supplementary Fig. 1a–e). Both 2nd-generation CAR constructs with a-EGFR nanobody (7D12), CD8α transmembrane domain, and with either 4-1BBζ or CD28ζ costimulatory domain were transduced via lentiviral vectors into human CD4⁺ or CD8⁺ T cells separately isolated from the peripheral blood of healthy donors after CD3/28-Dynabeads activation (Fig. 1a). The CAR expression was controlled under a constitutive SFFV promoter. Resting T cells from CD3/28-Dynabeads activation and CAR transduction were stimulated with CAREp at 2.5 particles per cell with the supplementation of 30 IU/mL IL-2 in the medium and were maintained at a concentration between 0.5 and 1.5 million per mL during the in vitro culture. Surface-presented EGFR ligands on CAREp progressively decreased during co-culture with CAR-T cells, with more than 50% signal reduction observed within 3 days (Supplementary Fig. 1f–h). Across three donors, CAREp stimulation induced a bell-shaped response pattern for cell proliferation with respect to the EGFR:a-CD28 Ab ligand ratio. The [9:1] ratio consistently yielded the highest cell expansion, while higher or lower proportions of a-CD28-Ab (e.g., [1:3] or [1:0]) led to reduced proliferation (Supplementary Fig. 2a). The trend was reproducible across donors and between IL-2 and IL-7/15 supplementations, despite inter-donor variation in absolute cell numbers (Supplementary Fig. 2a). Meanwhile, CAREp-induced cell proliferation required cytokine supplementation (Supplementary Fig. 2b). Alternative supplementation with IL-7 and IL-15 (5 ng/mL for each) also supported proliferation but was less effective than IL-2 (Supplementary Fig. 2a).

Using CAREp-[9:1] with 30 IU/mL IL-2 supplemented, we next evaluated repeated stimulation of CAR-T cells in vitro, with resting periods between stimulations. Remarkably, CD8⁺ 4-1BBζ CAR-T cells exhibited sustained proliferation for up to 100 days across three donors (Fig. 1c). In contrast, CAREp-[1:0] without a-CD28 Ab failed to support prolonged expansion (Supplementary Fig. 2c), highlighting the necessity of dual-signal input. Notably, stimulation every 4 days did not yield expansion, whereas an ~8-day interval allowing cells to rest from the prior stimulation was required to maintain proliferative capacity (Supplementary Fig. 2d). Moreover, resting cells gradually lost viability over ~7 days without re-stimulation, underscoring their reliance on continuous, antigen-driven signaling. Importantly, only CAR-transduced (EGFR-specific) T cells expanded in response to CAREp, while CAR-negative cells failed to proliferate and likely dwindled (Fig. 1d), confirming antigen specificity. Expansion was independent of CAR expression levels, indicating robust signaling irrespective of surface density (Fig. 1d). This trend was also observed with the CD28ζ construct, where cumulative expansion reached about 10⁴-fold after three CAREp stimulations and 10⁸-fold after six rounds (Fig. 1e). Although donor-to-donor variability is a common feature in human T cell studies^6,8, CAREp stimulation appeared to unify proliferative behavior across individuals (Fig. 1c). Depending on construct and donor, proliferation sustainability ranged from 90 to 130 days, with cumulative expansion from ~2³⁰ (10⁹) to 2⁶⁰ (10¹⁸)-fold (Fig. 1c and Supplementary Fig. 2e). These values refer to fold change relative to the starting population at the first CAREp stimulation. When including the preceding 100-1000-fold expansion from standard engineering steps (CD3/28-Dynabeads stimulation and CAR transduction, Fig. 1a), the data emphasize the extraordinary proliferative capacity of CAR-T cells when exposed to optimally structured signals. Interestingly, although CD28ζ-CAR showed a similar pattern of cell expansion after one round of stimulation using particles with different ligand ratios (Supplementary Fig. 2f), the additional amount of a-CD28 Ab on CAREp-[9:1] compared to CAREp-[1:0], ideal for 4-1BBζ-CAR, was not necessary for CD28ζ-CAR proliferation (Supplementary Fig. 2g). This is possibly due to the already incorporated CD28 costimulatory signaling domain in the CAR construct. However, CD28ζ-CAR T cells from the same set of donors showed heterogeneous expansion patterns during repeated stimulations with CAREp-[1:0] (Supplementary Fig. 2e). For donor 3 that did not respond well to CAREp-[1:0] and a fourth donor (donor 4), a low level of a-CD28 Ab on CAREp-[9:1] enhanced and sustained cell proliferation (Supplementary Fig. 2h). Additionally, CD4⁺ T cells showed similar robust proliferation by repeated CAREp stimulations (Supplementary Fig. 2i). Importantly, CAREp-induced expansion was dependent on CAR-antigen-specific stimulation, with minimal contribution from the chemical components of PLGA or DNA scaffold (Supplementary Fig. 2j,k). Furthermore, sustained proliferation was reproducibly observed in a-BCMA-CAR T cells generated using the clinically approved Abecma (idecabtagene vicleucel) construct containing a BB21 scFv, CD8α transmembrane domain, CD3ζ signaling domain, and 4-1BBζ costimulatory domain (Supplementary Fig. 2l).

Next, we asked whether the sustained T cell proliferation observed with CAREp could be achieved by other stimulation approaches, including direct CAR engagement by antigen-positive tumor cells, and commercially available Dynabeads via CD3/CD28 stimulation^11,25,31. To account for potential overstimulation effects (Supplementary Fig. 2d), T cells were re-stimulated only after a rest phase, mirroring the CAREp protocol. When 4-1BBζ-CAR-T cells were co-cultured with EGFR⁺ glioblastoma U87 cells (1:1 ratio), U87 cells were rapidly eliminated within 2-3 days, followed by T cell expansion for 5-7 days. However, proliferation failed to sustain beyond 3-4 stimulation rounds across all three donors tested (Fig. 1f), indicating premature exhaustion or cell death under repeated natural antigen exposure. CD28ζ-CAR T cells from donor 1 maintained proliferation for up to 60 days, but cells from other donors declined after ~20 days (Supplementary Fig. 2m). To further assess CAR-stimulation in a different cellular context, CD28ζ CAR-T cells were stimulated with EGFR-overexpressing K562 leukemia cells, which have previously been used as feeder cells for T cell expansion^11,12,32. Strikingly, even a single K562 stimulation induced high expression of PD-1, LAG-3, and TIM-3, along with increased Caspase 3/7-activity, in contrast to CAREp-expanded cells (Supplementary Fig. 2n). We then evaluated CD3/28-Dynabeads in the same rest-stimulate assay. Proliferation varied markedly among donors, and overall expansion after six rounds was orders of magnitude lower than with CAREp stimulation (Fig. 1g, Supplementary Fig. 2l). Furthermore, Dynabeads do not selectively stimulate CAR⁺ T cells, leading to population dilution. In some donors, CAR⁺ cells were entirely lost (Fig. 1g), indicating that standard CD3/28-Dynabeads activation not only underperforms but also expands non-transduced bystanders. Finally, to determine whether the observed proliferative advantage depended on the CAREp architecture itself, we compared CAREp with similarly sized magnetic particles functionalized with EGFR as well as DNA-scaffolded PLGA microparticles functionalized with a-CD3/CD28 antibodies. Although CD3/CD28-functionalized PLGA particles supported greater expansion than magnetic EGFR particles or Dynabeads, none of these alternative formulations achieved the level of CAR-T cell expansion induced by CAREp after three rounds of stimulation (Supplementary Fig. 2o). Together, these findings highlight the importance of the specific extracellular signal organization presented by CAREp in supporting sustained CAR-T cell proliferation.

Preservation of progenitor-like phenotype during CAREp-mediated expansion

To evaluate how repeated CAREp stimulation affects T cell differentiation, we profiled CAR-T cell phenotype across multiple stimulation rounds using flow cytometry. An initial 10-color panel measured canonical markers of memory and exhaustion, including CD62L, CD45RA, CCR7, CD127, PD-1, LAG-3, and TIM-3. Consistent with proliferation patterns (Fig. 2a), CAREp-[9:1] induced the highest percentage of CD62L⁺CD45RA⁺ cells across donors, outperforming suboptimal ratios such as [1:0] or [1:3] (Fig. 2b,c). These cells are broadly defined as memory progenitors, subsets associated with enhanced persistence and clinical efficacy^33–38. For both 4-1BBζ and CD28ζ CAR constructs, the memory progenitor population (CD62L⁺CD127⁺ or CD62L⁺CD45RA⁺) remained stable across up to six stimulations using optimally configured CAREp. Remarkably, even after expansion on the order of 10⁶–10¹²-fold, progenitor-like cells were maintained at ~20% in 4-1BBζ CAR-T and ~60% in CD28ζ CAR-T, comparable to the baseline prior to CAREp stimulation (UnS) (Fig. 2d, e). This population declined only after eight or more stimulation rounds (CAREp-8 or CAREp-9, Fig. 2d, e), suggesting a late-stage shift toward terminal differentiation.

Fig. 2. Memory-like progenitors are better preserved during repeated stimulations by CAREp versus CD3/28-Dynabeads.

a Relative expansion of CD8⁺ 4-1BBζ CAR-T cells 10 days after the first stimulation by CAREp coated with varying densities of EGFR-antigens and a-CD28 agonist antibodies (a-CD28 Ab) ([1:0], [9:1], [3:1], [1:3], [0:1]) in the presence of 30 IU/mL IL-2. UnS: unstimulated cells. b Representative CD62L and CD45RA expression of the expanded CD8⁺ T cells at 10 days after the first CAREp ([1:0], [9:1], [1:3]) stimulation. c CD45RA⁺CD62L⁺ population in CAREp ([1:0], [9:1], [1:3])-expanded cells in the presence of 30 IU/mL IL-2 or IL-7/15 (5 ng/mL each). Data are from n = 6 independent samples, and P values were determined by one-way ANOVA with Tukey’s multiple comparisons test. d Representative flow plots of CD127 and CD62L expression on CD8⁺ 4-1BBζ and CD28ζ CAR T cells that experienced 0-9 rounds of CAREp stimulation. e CD62L⁺CD127⁺ population of CD8⁺ EGFR-CAR T cells (4-1BBζ and CD28ζ constructs) before (UnS) and after 5-6 rounds of CAREp stimulation. Data represent mean of n = 3 donors, and P values were determined by two-way ANOVA with Sidak’s multiple comparisons. f UMAP and key marker expression of the pooled CD8⁺ EGFR-CAR T cell samples from repeated CAREp or CD3/28-Dynabeads stimulations after the immunostaining using a 25-marker spectrum flow panel. g Distribution of the expanded cells (with expansion fold in the bracelet) by Dynabeads or CAREp stimulations on the pooled UMAP in (f). h Cluster-1 population in CAREp or CD3/28-Dynabeads stimulated 4-1BBζ or CD28ζ CAR T cells with equivalent folds of cell expansion. Data are from n = 3 donors, and P value was determined by two-tailed paired t-test. Samples in (a–h) are from n = 3 healthy donors.

To gain a deeper insight into cell phenotypic changes, we applied spectrum flow cytometry using a 25-marker panel including CD45RA, CD62L, TCF1, CD127, CD27, CD28, CCR7, CD45RO, CD95, CD25, Ki67, ID2, CXCR3, HELIOS, IRF4, PRDM1, HLA-DR, KLRG1, TOX, CD244, CD39, CD38, LAG3, PD-1, and TIM3. Cells from multiple CAREp and CD3/28-Dynabeads stimulations were pooled and visualized using a UMAP projection (Fig. 2f and Supplementary Fig. 3a). Due to inconsistent expansion with Dynabeads, only cells from ≤5 rounds could be included. Manual gating based on memory- and exhaustion-associated markers resolved four clusters. Cluster-1, characterized by high CD62L, CD27, CD45RA, and CD127 and low PD-1, TIM-3, LAG-3, and CD39, was defined as a memory progenitor-like state (Fig. 2f and Supplementary Fig. 3b). We observed a gradual phenotypic progression from cluster-1 to more differentiated states with repeated CAREp stimulation, with divergence in trajectories between the two CAR constructs (Fig. 2g and Supplementary Fig. 3c–f). Importantly, cluster-1 remained preserved through 3–5 CAREp rounds, consistent with 10-color panel data (Fig. 2d, e). When comparing CAREp and Dynabeads stimulations matched for expansion fold, cluster-1 frequencies were consistently more than two folds higher with CAREp in both CAR constructs (Fig. 2g, h and Supplementary Fig. 3c, e, g, h), suggesting a quicker differentiation induced by Dynabeads. Collectively, these results demonstrate that CAREp stimulation enables massive expansion while delaying terminal differentiation, maintaining a progenitor-like phenotype through the early-to-mid phases of in vitro proliferation (up to rounds 5-6), during which ~10⁴–10⁸-fold expansion was achieved.

Rewired T cell signaling supports proliferation, telomere stability, and mitochondrial fitness

When T cells encounter antigen-expressing cancer targets, canonical signaling pathways, including calcineurin-NFAT, PKCθ-NF-κB, Ras-Raf-Erk, and chemotaxis pathways, are activated to mediate functions such as proliferation, migration, adhesion, and cytotoxicity³⁹. However, in CAR-engineered T cells, chronic antigen exposure often leads to rapid exhaustion and functional decline^4,25. To investigate how signaling profiles differ between stimulatory formats that do or do not support long-term CAR-T proliferation, we stimulated CD28ζ CAR-T cells from three donors with either CAREp-[1:0] or EGFR⁺ U87 tumor cells and performed bulk RNA sequencing at 24 h. Unstimulated CAR-T cells (UnS) served as controls. For U87 co-cultures, CD8⁺ T cells were FACS-purified to exclude tumor cell transcripts. Principal component (PC) analysis revealed that CAREp and U87 induced distinct transcriptomic programs that overrode inter-donor variation (Fig. 3a and Supplementary Fig. 4a). CAREp stimulation upregulated pro-survival and metabolic genes such as CD27, BCL2L10, MT-RNR1 and CYP1B1, while downregulating effector and exhaustion-associated transcripts, including PDCD1, LAG3, TNFRSF18, IL23A, and IL12RB2 (Supplementary Fig. 4b). Gene ontology (GO) analysis further revealed that pathways associated with apoptosis, TGF-β signaling, inflammatory response, actin cytoskeleton, migration, and chemotaxis were preferentially enriched following U87 stimulation, whereas CAREp stimulation preferentially enriched pathways associated with DNA repair, cell cycle, cytoplasmic translation, chromatin remodeling, telomere maintenance, mitochondrial function, and oxidative stress response (Supplementary Fig. 4c).

Fig. 3. CAREp rewires early signaling programs aligned with durable proliferation, telomere maintenance, and mitochondrial fitness.

a Principal component analysis of CD8⁺ CD28ζ CAR-T cell transcriptome by bulk RNA sequencing at 24 h after CAREp or U87 stimulation or without stimulation (UnS) among n = 3 donors (D1-3). b GSVA of signaling pathways significantly different between CAREp- and U87-stimulated cells (adjusted P value < 0.01). c,d Telomere lengths of CD8⁺ CD28ζ CAR^-T cells after varying rounds of CAREp (c) or U87 stimulation (d) by the T/S assay. e Telomerase activity by the TRAP analysis of CD8⁺ CD28ζ CAR^-T cells at 3 days after CAREp-restimulation following varying rounds of CAREp-expansion. f Telomerase activity of CD8⁺ CD28ζ CAR^-T cells at 3 days after CAREp versus U87 stimulation. Donor 1 was measured at the second round of CAREp or U87 stimulation, and donors 2 and 3 were measured at the first round. g Telomerase activity at 0 and 3 days after restimulation of CD8⁺ CD28ζ CAR^-T cells that experienced 1-round of CAREp or U87 stimulation. Data in (c–g) represent mean ± SD from n = 3 technical replicates, and P values were determined by one-way ANOVA with Tukey’s multiple comparisons test (c), two-way ANOVA with Sidak’s multiple comparisons test (d), two-way ANOVA with Tukey’s multiple comparison test (f), and two-tailed unpaired ttest (g). h Representative oxygen consumption rate (OCR) of CAREp-expanded CD8⁺ 4-1BBζ and CD28ζ CAR-T cells from Seahorse Mitostress assay. Data are mean ± SD from n = 6 biological replicates. i Spare respiration capacity (SRC) of CAREp-expanded CD8⁺ 4-1BBζ and CD28ζ CAR-T cells from Seahorse Mitostress assay. Data are mean ± SD from n = 6 biological replicates, P values were determined for each donor individually using one-way ANOVA with Tukey’s multiple comparisons test for comparisons among three or more timepoints, or a two-tailed unpaired t test for comparisons between two timepoints. Statistical significance is represented by exact P values or P < 0.0001, as indicated in the figure panels.

To further refine these signaling differences, we first performed gene set enrichment analysis (GSEA) between CAREp- and U87-stimulated cells to identify differentially regulated pathways, followed by gene set variation analysis (GSVA) of selected pathways across all samples (Fig. 3b and Supplementary Fig. 4d). CAREp consistently enriched pathways associated with DNA repair, chromatin remodeling, telomere maintenance, and mitochondrial function, while comparatively restraining cytotoxic, chemotactic, and differentiation-associated signaling programs, including Calcineurin-NFAT, PKCθ-NF-κB, and JAK-STAT pathways relative to U87-stimulated cells (Fig. 3b). Phospho-flow analysis further revealed distinct acute signaling dynamics between CAREp and U87 stimulation. CAREp stimulation induced elevated ERK1/2 phosphorylation (pT202/pY204) together with comparatively reduced PLCγ phosphorylation, whereas U87 stimulation showed a different phosphorylation profile (Supplementary Fig. 4e). Together, these findings suggest that CAREp induces a distinct signaling program characterized by coordinated proliferative and metabolic programming while limiting differentiation-associated signaling outputs.

Given the prominent enrichment of telomere maintenance and DNA repair pathways in CAREp-stimulated cells, we next asked whether these signaling differences translated into altered telomere dynamics during repeated expansion. Telomere length was measured across multiple rounds of CAREp stimulation using quantitative PCR (T/S assay)⁴⁰ (Fig. 3c). Like other physiological cells, continuous expansion gradually reduced telomere length across all three donors (Fig. 3c and Supplementary Fig. 5a, b). However, telomere length remained stable during the first three rounds of CAREp stimulation (approximately 20 cell divisions) (Fig. 3c, d and Supplementary Fig. 5c). Among the three donors, donor 1 exhibited the greatest proliferative sustainability under U87 repeated stimulation (Supplementary Fig.2m), enabling sufficient recovery of cells for telomere analysis after three stimulation rounds. Besides less overall expansion fold (~10 divisions by U87 vs 20 divisions by CAREp), cells yielded from three U87 stimulations showed significantly decreased telomere length compared to one round of stimulation, whereas telomere length stayed the same between one and three rounds of CAREp stimulation (Fig. 3d). This further indicates that CAREp stimulation preserves telomere ends during early massive expansion, whereas accelerated telomere erosion happens with cancer-stimulated cell expansion.

To maintain self-renewal capacity and super longevity, memory T cells can upregulate their telomerase activity upon activation to compensate for telomere shortening during cell proliferation^19–21. To investigate the telomerase activity after CAREp stimulations, frozen-stocked resting cells from multiple rounds of stimulation were thawed, stimulated with CAREp, and analyzed after three days using the telomerase repeated amplification protocol (TRAP)⁴¹. Resting cells displayed negligible telomerase activity (Supplementary Fig. 5d), whereas early CAREp-expanded cells (rounds 0-2) showed strong telomerase induction (Fig. 3e). Notably, this inducibility progressively declined during later rounds of stimulation, consistent with gradual loss of stem-like properties. In contrast, U87-stimulated CAR-T cells exhibited significantly lower telomerase activity across all donors (Fig. 3f). Consistent with the differential telomere retention observed above (Fig. 3d) a second round of CAREp stimulation induced substantially higher telomerase activity than a second round of U87 stimulation (Fig. 3g).

Given the enrichment of mitochondrial pathways in CAREp-stimulated cells (Fig. 3b), we next assessed mitochondrial fitness of expanded cells using the Seahorse MitoStress assay^42,43. Rested CD8⁺ CAR-T cells following repeated CAREp stimulation were sequentially treated with respiration inhibitors, including oligomycin, FCCP, and rotenone/antimycin A, to measure oxygen consumption rate (OCR) as an indicator of respiratory function (Fig. 3h). For both 4-1BBζ-CAR and CD28ζ-CAR constructs, cells expanded through 3-6 rounds of CAREp stimulation retained or improved respiratory capacity compared to baseline (Fig. 3h and Supplementary Fig. 5e). Across multiple donors, we observed a consistent increase in spare respiratory capacity (SRC), supporting enhanced mitochondrial fitness following CAREp-mediated expansion (Fig. 3i).

Together, these findings demonstrate that CAREp induces an early signaling and transcriptional state that closely aligns with long-term functional outcomes. CAREp stimulation coordinated transcriptional programs associated with DNA repair, chromatin remodeling, telomere maintenance, and mitochondrial fitness, while comparatively limiting differentiation- and exhaustion-associated signaling. These early programs corresponded with sustained proliferation, transient telomerase induction, delayed telomere attrition, enhanced respiratory capacity, and preservation of memory-associated states during prolonged expansion. Collectively, these data support a model in which precisely structured extracellular cues can synchronize acute CAR-T signaling dynamics with durable proliferative and metabolic fate programs.

Well retained effector function of CAREp-expanded CAR-T cells

To assess effector function, we performed tumor cell killing assays using K562 cells engineered to express truncated EGFR. This target was relatively resistant to CAR-T cytotoxicity, particularly for CD28ζ CAR-T cells, which underwent apoptosis upon co-culture and failed to eliminate the target (Supplementary Fig. 2n). In contrast, 4-1BBζ CAR-T cells exhibited modest killing ability, which was markedly enhanced after three rounds of CAREp stimulations (Fig. 4a, b). These expanded cells also persisted at higher numbers following a 4-day co-culture (Fig. 4a). However, this cytotoxic advantage diminished after 7-8 stimulation rounds (Fig. 4c), indicating a functional ceiling after extended expansion. TNF-α secretion patterns mirrored killing efficacy, peaking after three CAREp rounds and declining after extensive expansion (Fig. 4d).

Fig. 4. Massively expanded cells demonstrate robust effector function.

a Flow cytometry plot of EGFR-overexpressing K562 cells at 4 days after the co-culture with CD8⁺ 4-1BBζ CAR-T cells with or without CAREp-stimulation. E:T: effector-to-target ratio. b Target killing efficacy (%) between unstimulated CAR-T cells and 3-rounds of CAREp-stimulated cells in (a). Data are from n = 3 donors, and P value was determined by two-tailed paired ttest. c,d Target killing efficacy (%) at day-4 (c) and supernatant TNF-α level at day-2 (d) after the co-culture of K562 and T cells experienced 0-8 rounds of CAREp-stimulation. Data are mean ± SD from n = 3 biological replicates. e Growth of U87 cell line co-cultured with CD8⁺ 4-1BBζ CAR-T cells expanded after 0 (UnS), 3, and 8 rounds of CAREp stimulation. Data are mean ± SD from n = 4 biological replicates. f Target killing efficacy (%) of CAREp-expanded CD8⁺ 4-1BBζ and CD28ζ CAR-T cells against U87 cell line at 72 h after the co-culture. Data are mean from n = 4 biological replicates. g Serial killing of CAREp-expanded CD8⁺ 4-1BBζ CAR-T cells against U87 cells for three rounds. Data are mean from n = 4 biological replicates. h Schematic of the in vivo experimental design evaluating anti-tumor activity of CAREp-expanded CAR-T cells in NSG mice bearing subcutaneous U87 tumors. Mice received peri-/intratumoral injection of 0.5 × 10⁶ CD4⁺ and 0.5 × 10⁶ CD8⁺ CD28ζ CAR-T cells two weeks after tumor implantation. i Survival curves of NSG mice bearing U87 tumors following peri-/intratumoral injection of PBS, unstimulated cells (UnS), one round of Dynabeads-expanded cells (Dynabeads-1), one round of CAREp-expanded cells (CAREp-1), or three rounds of CAREp-expanded cells (CAREp-3). n = 5 mice per group. Statistical significance was determined using the Mantel-Cox test with Holm-Sidak’s multiple comparisons test. j Tumor growth curves of NSG mice groups in i. Data are presented as mean ± s.e.m. from n = 5 mice per group, and P values were determined using one-way ANOVA with Tukey’s multiple comparisons test. Statistical significance is represented by exact P values or P > 0.9999, as indicated in the figure panels.

To evaluate killing of a more physiologically relevant target, we co-cultured CAREp-expanded CAR-T cells with EGFR⁺ U87 glioblastoma cells at an effector-to-target (E:T) ratio of 1:4 (Fig. 4e). Through an Incucyte imaging assay, CAREp-expanded cells (CAREp-3 and CAREp-8) exhibited more rapid killing kinetics compared to unstimulated controls (UnS) (Fig. 4f). Across all donors and both CAR constructs, T cells expanded for up to 11 CAREp rounds maintained potent cytotoxicity (Fig. 4f). Beyond 11 rounds, efficacy declined, consistent with the expansion plateau. For 4-1BBζ CAR-T cells, three rounds of CAREp expansion yielded cells with improved cytotoxicity, similar to the trend observed in K562 killing (Fig. 4f, g). Cytokine secretion during U87 killing was also profiled (Supplementary Fig. 6a–d). CAREp-expanded T cells produced similar levels of TNF-α and IL-2 as unstimulated controls at 48 h post co-culture (E:T = 1:1), but secreted lower levels of IFN-γ (Supplementary Fig. 6b, d). Finally, we evaluated serial killing capacity in repeated co-culture cycles at E:T = 1:4. CAR-T cells expanded through up to 8 rounds of CAREp maintained robust serial killing over three to four tumor challenges (Fig. 4g and Supplementary Fig. 6e). Given that extensively stimulated cells showed potent cytotoxicity and serial killing efficacy, a lower level of IFN-γ secretion may be due to the cell population evolution with contracted short-lived effector cells and enriched memory cells (Fig. 2). Notably, IFN-γ signaling is dispensable, and potentially inhibitory, for CD8⁺ T cell memory formation^44–46.

To further evaluate whether CAREp-expanded cells retained anti-tumor activity in vivo, we next tested CAR-T function in NSG mice bearing xenograft tumors. However, systemic delivery of the EGFR-CAR construct used in this study failed to effectively control tumor growth in vivo, whereas a CD19-CAR control efficiently eliminated Nalm6-EGFR tumors (Supplementary Fig. 7a), likely due to the off-tumor reactivity to bystander tissues^47,48. To overcome this limitation and enable initial assessment of in vivo function, CAR-T cells were administered through peri-/intratumoral injection^47,48 at a total dose of 1 million cells (0.5 million CD4⁺ plus 0.5 million CD8⁺) two weeks after subcutaneous U87 tumor implantation (Fig. 4h and Supplementary Fig. 7b). Under these conditions, CD28ζ CAR-T cells exhibited greater anti-tumor efficacy than 4-1BBζ CAR-T cells (Supplementary Fig. 7b).

We next compared CD28ζ CAR-T cells from different ex vivo expansion conditions, including unstimulated cells (UnS, conventionally engineered CAR-T cells recovered for one day post-thawing), one round of Dynabeads stimulation (Dynabeads-1), and one or three rounds of CAREp stimulation (CAREp -1 and -3), using equivalent cell doses. CAREp-1, CAREp-3, and Dynabeads-1 groups all retained potent anti-tumor activity comparable to unstimulated CAR-T cells (Fig. 4i, j and Supplementary Fig. 7c). Notably, one mouse in the Dynabeads-1 group and three mice in the CAREp-3 group exhibited tumor relapse after day 50 (Supplementary Fig. 7c), potentially reflecting the progressive effector differentiation observed following extended stimulation with this donor (Supplementary Fig. 7d). Consistent with this interpretation, one round of Dynabeads stimulation induced minimal cell expansion (~2 folds) but rapidly promoted effector differentiation, depletion of stem-like progenitors, and elevated PD-1 and TIM-3 expression relative to CAREp-expanded cells (Supplementary Fig. 7d). Together, these findings indicate that CAREp-expanded CAR-T cells retain substantial anti-tumor function in vivo while highlighting the importance of balancing prolonged expansion with preservation of durable progenitor-associated states.

Clonal enrichment of memory progenitor and durable effector states

To dissect the phenotypic and clonal trajectories underlying CAREp-induced expansion, we performed single-cell RNA sequencing (scRNA-seq) with paired CITE-seq and TCR tracking on CD28ζ CAR-T cells across four timepoints per donor (n = 3 donors; 12 total samples; Fig. 5a). Resting cells were barcoded with hashtags, pooled, and stained with a 130-marker CITE panel prior to droplet-based sequencing. This strategy allowed for integrated transcriptomic, surface protein, and clonal analysis while minimizing batch effects (Fig. 5a and Supplementary Fig. 8a). After data quality control, robust numbers of cells were identified among the three libraries, although cells from donor 3 after 5 and 9 rounds of CAREp stimulation showed less robust quality after the freeze-thaw cycle during sample preparation (Supplementary Fig. 8a).

Fig. 5. CAREp stimulation generates two transcriptionally distinct T cell subpopulations with memory progenitor and persistent effector-like states.

a Schematic of single cell RNA sequencing of resting CD8⁺ CD28ζ CAR-T cells from 3 donors x 4 timepoints with simultaneous analysis of transcriptome (GEX), 130 surface markers (CITE), and TCR repertoire (TCR). Longitudinal expansion curves of the donors are included in Supplementary Fig. 2e. 12 samples (5000 cells each for UnS, 3000 each for other timepoints) were individually stained with hashtag and pooled together for sequencing. (b) UMAP of GEX data highlighting cells from 0, 3, 6, 9 rounds of CAREp stimulation (donor 1). Light grey dots represent cells from all samples. c,d Unsupervised leiden clustering of GEX data and differentially expressed genes in the leidens. e Enrichment score of leiden cluster 3 vs 5 and 3 vs 1 from GSE9650 gene set enrichment analysis (GSEA). NES: normalized enrichment score; FDR: false discovery rate. f Trajectory of cells by RNA velocity analysis of GEX data. g Cell cycle distribution (G1, G2M, and S) in all sample UMAP from cell cycle analysis. h Relative RNA expression of selected genes on the GEX-UMAP. i Relative expression of selected surface markers (protein) in leiden 1, 5, and 6. Markers in orange, blue, and black are respectively associated with memory, effector, and exhaustion phenotypes. j,k Gene set scoring (Scanpy) of all sample GEX data using reported gene sets that showed correlation with patient complete remission (CR) or nonresponding (NR)^14,53.

UMAP projection and Leiden clustering of gene expression data revealed six distinct clusters with gene signatures for memory and effector phenotypes^49,50 (Fig. 5b–d and Supplementary Fig. 8b). Clusters 1-4 contained mostly unstimulated cells; clusters 5 and 6 were enriched with CAREp-expanded cells (Fig. 5b and Supplementary Fig. 8b). Cluster 1 expressed canonical memory progenitor markers (e.g., IL7R, CD27, LEF1, SELL, CCR7, and TCF7), while cluster 3 and cluster 4 respectively showed effector-associated genes (e.g., CCL5, GZMA) and terminal differentiated genes (e.g., TOX, BATF), reflecting conventional T cell differentiation (Fig. 5c–e and Supplementary Fig. 8b–g). A high similarity was observed between cluster 1 and 5, while cluster 6 showed dramatic difference from these two clusters (Fig. 5c). Cluster 5 showed decreased expression of memory-associated genes (TCF7, CD27, LEF1) and increased expression of effector-associated genes (CCL5, GZMA) compared to cluster 1, suggesting a differentiation shift to the effector-like state. Yet, gene set enrichment analysis (GSEA) revealed that cluster 5 still represented a higher level of progenitor state than cluster 3 (Fig. 5e). Consistently, RNA velocity analysis showed two distinct trajectories of cell differentiation: one from cluster 1 to 3 to 4 (grey), and the other one from cluster 1 to 5 to 6 (magenta), aligning well with the GSEA results (Fig. 5e, f). Thus, CAREp stimulation bifurcated the differentiation trajectory: instead of all cells progressing from Memory (cluster 1) → Effector (cluster 3) → Terminal (cluster 4) as in conventional differentiation, many CAREp-expanded cells entered an alternative route: Memory (cluster 1) → Effector-memory hybrid (cluster 5) → Durable effector-like (cluster 6) (Fig. 5f). Interestingly, cell cycle analysis showed that cells at clusters 1, 2, 5, and 6 primarily reside at G1 phase, while cluster 3 and 4 reside at G2M and S phases (Fig. 5g), responding to the acutely regulated cell cycle pathways by CAREp stimulation as well (Fig. 3b).

Compared with cluster 1 for both RNA and protein expression, cluster 5 maintained moderate levels of memory markers SELL/CD62L (both RNA and protein) and IL7R/CD127 (RNA), upregulated effector marker IL2RA/CD25 (both RNA and protein), and maintained low levels of senescence or exhaustion markers (PDCD1/CD279/PD-1, LAG3, HAVCR2/TIM3, KLRG1, TIGIT, both RNA and protein), indicating an effector-memory state (Fig. 5h, i and Supplementary Fig. 8d–h). Compared with cluster 5, cluster 6 showed a decrease of SELL/CD62L (both RNA and protein) and IL7R/CD127 (RNA), an increase of effector markers including CD69, CD244, CX3CR1, CD122, CD45RO (protein), and low levels of PDCD1, LAG3, HAVCR2/TIM3, TIGIT, and CD39 (RNA) (Fig. 5h, i and Supplementary Fig. 8d–h). Although cluster 6 is transcriptomically similar to a terminal effector (like cluster 4) (Fig. 5c), it lacked surface expression of KLRG1 and TIGIT, and instead showed persistent CD25, CD122, and STAT5 activation (Fig. 5h, i and Supplementary Fig.8i). Therefore, cells in cluster 6 clearly resemble the durable effector-like state during chronic antigen exposure^51,52. From CAR-T cell clinical trials, several gene sets in manufactured cell products have been identified to importantly correlate with patient remission or non-responding^14,53. To further assess the cell quality yielded from CAREp-stimulations, we carried out gene set scoring (Scanpy) of three gene sets including FOXO1 regulon¹⁴, complete remission (CR), and nonresponding (NR) in the GEX data set⁵³. Interestingly, cluster 5 and 6 showed high scores of FOXO1 regulon and CR gene sets and low scores of NR gene set, comparative to memory progenitor cells in cluster 1 (Fig. 5j, k).

From TCR repertoire analysis, we identified a clear trend of clonal enrichment over time among all donors, although that CAREp stimulation is TCR-independent (Fig. 6a). Due to suboptimal gene counts in clusters 4 and 6, TCR repertoire was more successfully tracked in clusters 1–3 and 5 (Supplementary Fig. 8c). Still, clones with drastically different lifespan and proliferative timeline were identified (Fig. 6b). For example, some clones (e.g., clones 4 and 28) vanished by day ~30; some clones (e.g., clone 43, 89, and 39) actively proliferate for 60-70 days and stabilized after 90 days; and some other clones (e.g., clones 14, 47, and 19) actively proliferate beyond 90-100 days (Fig. 6b). We then sorted clones with similar patterns of lifespan to three groups (G1—untraceable after 30 days, G2—untraceable after 70 days, and G3—traceable at 90-100 days) to track their phenotypic evolution (Fig. 6c). Notably, clones destined to fade early (G1) showed an overlap of memory (CD127, CD62L, and CD27) alongside effector/exhaustion (CX3CR1, KLRG1, CD244, TIGIT, CD279, CD152, CD223) features at the gene/protein level (Fig. 6d–f). In contrast, the longer-lived clones (G2 and G3) largely lacked these exhaustion markers from the start, suggesting a more pristine progenitor state (Fig. 6d–f). Similarly, between G2 and G3 and among clones of different lifespan, the expression of effector and exhaustion associated markers (CCL5, TOX, CD38, CD25, CD45RO) as well as the levels of T cell activation processes at a similar timeframe were relatively lower in longer-living clones (Fig. 6d–f and Supplementary Fig. 9a–d).

Fig. 6. Clonal tracking reveals heterogeneous proliferative lifespans, with CAREp-expanded long-lived clones maintaining stable, progenitor-like states.

a Clonal distribution of CD8⁺ CD28ζ CAR-T cells after the expansion from varying rounds of CAREp-stimulation by TCR repertoire analysis. b Dominant clones with more than 200 cells and their cell number distribution at different timepoints. c UMAP locations of traceable clones (clone size > 50 cells) with different proliferation capacity: group 1 (G1) − untraceable at day-30-40; group 2 (G2) − traceable at day-60-70; group 3 (G3) − traceable at day-90-100. G3-1, G3-2, and G3-3 represent group 3 cells exist at day-30-40, day-60-70, and day-90-100, respectively. d–f Balloon plot showing expression of key markers associated with memory (green line), effector (blue line), and exhaustion (orange line) phenotypes in assorted groups in (c). d,e show RNA levels from GEX data, and f shows protein levels from CITE data. g Violin plots represent GSVA-CR and NR scores of leidens 1-3 (Fig. 5d) and G3–1. Data shows the median value, along with the 25 and 75th quartiles; n represents the number of single cells analyzed in each cluster (CR and NR), n = 3374 (cluster1), 3911 (cluster2), 9016 (cluster3), and 543 (clusterG3-1). Medians were 0.1409, 0.1639, 0.0953, and 0.1436 for the CR signature, and −0.0347, −0.0362, 0.0062, and −0.0355 for the NR signature in clusters 1, 2, 3, and G3-1, respectively. h Violin plots represent GSVA-CR, NR, and FOXO1 regulon scores of assorted groups in (c). Data shows the median value, along with the 25th and 75th quartiles; n represents the number of single cells analyzed in each cluster (CR, NR, and FOXO1): G1, n = 1569; G2, n = 1357; G3-1, n = 543; G3-2, n = 2850; G3-3, n = 3414, Median CR scores were 0.1329, 0.1426, 0.1435, 0.1320, and 0.1218; median NR scores were −0.0189, −0.0281, −0.0355, −0.0239, and −0.0169; and median FOXO1 scores were −0.0370, −0.0290, −0.0236, −0.0282, and −0.0253 for G1, G2, G3-1, G3-2, and G3-3, respectively. P values in (g,h) are from One-way ANOVA with Tukey’s multiple comparisons. Statistical significance is represented by exact P values or P < 0.0001, as indicated in the figure panels.

G3 clones were then tracked at different timepoints, and a relatively stable immune profile with a gradual differentiation shift was observed along massive expansion (Fig. 6c). For example, RNA levels of FAS, PTGER2, BHLHE40⁴⁹, BATF, TOX, TIGIT, CTLA4, ENTPD1, and CD38, and protein levels of CD25, CD39, CD95 showed a gradual increase, while T cell proliferation transcription factor MYC (RNA) gradually decreased, and TNF (RNA), ITGB7 (RNA), CD69 (protein), CD134 (protein), CD45RO (protein), CD122 (protein), and CX3CR1 (protein) firstly increased then decreased (Fig. 6d–f). When we run gene set scoring of CR, NR, and FOXO1 regulon gene sets, G3 at 30-40 days (G3-1) showed the best score comparable to Leiden cluster-1 (Fig. 6g). However, cell quality indicated by CR and NR scores decreased over time (G3-2 and G3-3 compared to G3-1) (Fig. 6h). This suggests that by ~day 90-100 ( > 8 stimulations), even the most durable clones reach a natural limit, coinciding with the plateau of expansion and cytotoxicity observed earlier (Figs. 1, 2, 4). In other words, while CAREp delays the onset of terminal differentiation, it does not entirely prevent it—cells eventually approach their physiological lifespan. Additionally, CAR RNA expression spectrum among clones of different longevity did not show clear difference (Supplementary Fig. 9e, f). Altogether, these data demonstrate that CAREp stimulations selectively expanded durable clones and bifurcate cells to a distinct differentiation trajectory with massive proliferation and a sustained, less-exhausted phenotype, which may underlie their improved therapeutic potential.

Discussion

We achieved sustained in vitro expansion of human CD8⁺ CAR-T cells for over 100 days using biomaterial-mediated stimulation, revealing the remarkable proliferative capacity of human T cells under optimally structured extracellular cues^9,11,12. CAREp stimulation induced a distinct transcriptional program characterized by coordinated activation of pathways associated with DNA repair, chromatin remodeling, telomere maintenance, and mitochondrial fitness together with modulation of canonical differentiation-associated signaling pathways, including NFAT, NF-κB, JAK-STAT, and ERK-associated programs. Importantly, these early signaling and transcriptional states closely aligned with long-term functional outcomes, including sustained proliferation, delayed telomere attrition, preserved progenitor-like states, and enhanced metabolic fitness. Unlike natural tumor stimulation that tightly couples proliferation with exhaustion and differentiation, CAREp enabled prolonged cell division while delaying terminal dysfunction, suggesting that extracellular signal architecture can synchronize acute signaling dynamics with durable CAR-T cell fate programs.

Balanced delivery of CAR antigen and costimulatory signals enabled massive expansion not achievable with tumor-cell stimulation or CD3/CD28-Dynabeads, both of which induced more rapid differentiation or attrition. For 4-1BBζ-CARs, low-density a-CD28 costimulation ([9:1] ratio) was essential for sustained proliferation, whereas CD28ζ CAR-T cells required minimal or no external CD28 input in most donors. Importantly, alternative stimulation formats, including EGFR-overexpressing K562 feeder cells, magnetic EGFR particles, and CD3/CD28-functionalized PLGA particles, failed to reproduce the same degree of sustained proliferation. These observations underscore that the proliferative phenotype depends not simply on antigen stimulation alone, but on the precise extracellular signal architecture imposed by CAREp. In addition to chemical composition, DNA scaffolds provide nanoscale control over ligand spacing, rigidity, and flexibility, parameters^54,55 that may further influence CAR signaling dynamics and remain important areas for future investigation.

Although CAREp stimulation was independent of TCR engagement, clonal enrichment was consistently observed across donors, suggesting that the peripheral repertoire contains subsets with intrinsically greater proliferative potential. CAREp preferentially expanded these memory-associated clones while reducing donor-to-donor variability^6,8,34, indicating a platform capable of enriching high-fitness populations often lost under standard protocols. CAREp-expanded cells also exhibited distinct differentiation trajectories during chronic stimulation. Expanded cells maintained high CD25 expression and cytokine dependence while preserving low expression of exhaustion-associated markers, including PD-1, TIGIT, and KLRG1. In parallel, IFN-γ secretion remained comparatively restrained despite preserved cytotoxicity and TNF-α/IL-2 production. Given prior evidence that excessive IFN-γ signaling can impair CD8⁺ memory differentiation^44–46,56, this altered cytokine profile may contribute to sustained proliferative potential. Together with coordinated TOX and STAT5 expression, these findings suggest that CAREp promotes durable proliferation of effector-like cells during chronic stimulation⁵¹. Nevertheless, transcriptomic analyses after prolonged culture (~60-70 days) revealed gradual progression toward terminal differentiation states, including increased expression of FAS, ENTPD1, CTLA4, TIGIT, BHLHE40, BATF, and TOX together with reduced ITGB7, MYC, CCL5, and GZMA expression^49,52. Thus, CAREp delayed exhaustion-associated progression while sustaining proliferative capacity over extended periods.

Importantly, extensively expanded cells retained features of physiological regulation rather than transformation-like behavior. Proliferation ceased upon CAREp withdrawal, cytotoxicity gradually declined after extensive stimulation, and telomerase induction remained transient during early expansion phases. Correspondingly, telomere length was initially preserved but progressively shortened during prolonged culture, consistent with regulated cellular aging rather than uncontrolled immortalization. These observations suggest that CAREp does not abolish normal differentiation and proliferative limits but instead prolongs functional expansion while maintaining physiological control mechanisms.

These durable CAR-T cell states closely mirrored transcriptional signatures associated with clinical response^34,37,53. Given the established relationship between progenitor preservation and manufacturing quality, CAREp-mediated stimulation, particularly during early stimulation rounds, may provide a valuable strategy for expanding high-fitness CAR-T populations ex vivo while delaying differentiation-associated attrition^57,58. In this study, CAREp was designed to function downstream of conventional CAR engineering workflows by providing programmable CAR-specific stimulation after CAR expression had been established. Notably, our previous work demonstrated that CD3/CD28-functionalized DNA-scaffolded PLGA particles can replace Dynabeads for initial T cell activation prior to CAR transduction²⁷. Additionally, because PLGA is biocompatible and biodegradable, the platform may provide opportunities for programmable in vivo stimulation or post-infusion control of engineered T cells, an increasingly important direction given the growing interest in direct in vivo CAR engineering strategies.

The present study also highlights several translational considerations. Due to the off-tumor toxicity associated with EGFR-targeting CARs, in vivo studies in this work required peri-/intratumoral delivery to enable initial functional assessment. Ongoing work therefore focuses on evaluating clinically relevant CAR systems, including BCMA-targeting constructs, in representative tumor models. In addition, the current study primarily focused on CD8⁺ CAR-T cells, whereas coordinated optimization of CD4⁺ subsets essential for durable tumor control^59,60, freeze-thaw recovery protocols, differentiation balance, and cell dosing will likely further influence long-term therapeutic performance. Together, these findings establish CAREp not only as a biomaterial platform for sustained CAR-T expansion, but also as a programmable system for dissecting how nanoscale extracellular cue organization governs signaling dynamics, proliferative capacity, metabolic fitness, and differentiation trajectory in engineered T cells. More broadly, this work highlights how advances in precision biomaterial engineering can provide mechanistic control over immune-cell fate and function, thereby informing future strategies for both ex vivo manufacturing and in vivo immune programming.

Methods

Synthesis of DNA-scaffolded microparticles

DNA-scaffolded particles were synthesized as previously described^27,28. Briefly, polymer (PLGA)-DNA amphiphiles were generated and used as the sole surfactant for microparticle fabrication. All chemically modified DNA strands were custom synthesized and obtained from Integrated DNA Technologies. To prepare PLGA-DNA amphiphiles, 3′-thiol-modified DNA oligonucleotides (G and R, Supplementary Table 1) were deprotected by treatment with a 100-fold molar excess of tris(2-carboxyethyl)phosphine hydrochloride (TCEP; Sigma, catalogue no. 646547) at 37 °C for 1 h. The deprotected oligonucleotides were then purified using size-exclusion chromatography (Glen Research, catalogue no. 61-5010). Freshly prepared 17-mer thiol-DNA at 200 μM was reacted with poly(lactide-co-glycolide)-b-poly(ethylene glycol)-maleimide (PLGA10k-PEG5k-Mal; Akina, catalogue no. AI053) at a 1:1 molar ratio. The reaction was carried out in a solvent mixture of dimethylformamide (DMF)/H₂O (90:10, v/v) containing 0.2% triethanolamine and incubated overnight at room temperature. The following day, polymer-DNA strand G and strand R were mixed at a total amount of 100 nmol with the desired ratio. The solvent was removed by vacuum evaporation at 70 °C for 3 h, and the dried amphiphile mixture was stored at −20 °C until use.

To fabricate PLGA microparticles, the dried polymer-DNA reaction mixture was reconstituted in 200 μL of solvent composed of 100 μL ethyl acetate (EtOAc) and 100 μL H₂O. This solution was then added to a mixture containing 0.5 ml of EtOAc, 50 mg of unmodified polymer (poly(lactic-co-glycolic acid), 50:50, Mw = 38,000–54,000 Da; Sigma, catalogue no. 719900), and 0.5 ml of aqueous buffer (10 mM sodium citrate, 600 mM Na⁺, pH 3.0). The entire mixture was vortexed and probe-sonicated on ice (Fisher miniRoto, catalogue no. S56) at 7–8 W for five 5-second pulses with 10-second intervals. Immediately after sonication, the emulsion was added to 9 ml of 0.2% PVA and stirred under a chemical hood for 3 h to allow complete evaporation of ethyl acetate. The resulting microparticles were filtered through a 40-μm cell strainer (Sigma, catalogue no. CLS431750) and washed by centrifugation at 10,000 g for 10 min. The pellet was resuspended in TE buffer (10 mM Tris-HCl, pH 8.0) containing 0.01% Tween-20 (VWR, catalogue no. 97063-872). This washing protocol was consistently used for all subsequent particle fabrications. After three washes, particles were resuspended in TE buffer supplemented with 1% PVA (Bioworld, catalogue no. 41600201-2) and 0.1 M TEAA (Neta Scientific, catalogue no. 471283), then lyophilized using a VirTis Advantage Plus freeze dryer for long-term storage.

Attachment of complementary DNA to a-CD28 antibody and purification

Anti-human CD28 antibody (BioXCell, catalogue no. BE0248) was conjugated to the complementary DNA strand (CompR, Supplementary Table 1) through the “TCEP” strategy^27,28. Antibodies were first exchanged in reducing buffer (PBS with 10 mM EDTA) by size exclusion chromatography (Glen Gel-Pak 0.2 Desalting Column, catalogue no. 61-5002), and the disulfide bond at the hinge region was selectively reduced by adding TCEP with a 4.5 molar excess and incubating at 37 °C for 1 h. Excess TCEP was removed through size exclusion chromatography. 3’-NH2 modified complementary DNA (CompR) was conjugated to a MAL-dPEG4-NHS linker (BroadPharm, catalogue no. BP-20554) with 30-fold molar excess in HEPES buffer (pH 7.0) at 37 °C for 1 h, followed by the removal of the unconjugated linker through size exclusion chromatography and 70% ethanol precipitation. The reduced antibody and modified DNA were combined at a 1:4 molar ratio and incubated at 37 °C for 1 h, then 4 °C overnight. The next day, CompR-a-CD28 antibody conjugates were purified using protein G affinity chromatography (Genscript catalogue no. L00209) to remove unconjugated DNA. For surface quantification on particles, CompR-a-CD28 antibody was further fluorescently labeled through a reaction with NHS-AFDye647 (Click Chemistry Tools, catalogue no. 16820) with an 8× molar excess at 37 °C for 1 h, followed by purification using size-exclusion chromatography (Thermo Scientific catalogue no. 89883).

Surface functionalization of microparticles to generate CAREp

The surface functionalization of PLGA microparticles was performed as previously described^27,28. Briefly, particle concentration was normalized based on optical density at 550 nm (OD₅₅₀), measured using a UV-Vis spectrophotometer (Thermo Scientific NanoDrop One/OneC Microvolume UV-Vis Spectrophotometer). DNA-scaffolded PLGA microparticles were hybridized with biotinylated complementary DNA strands (CompG, Supplementary Table 1) plus CompR-antiCD28 at a specific molar excess to PLGA DNA scaffolded particle amount (Supplementary Table 2) by incubating the particles at 37 °C for 30 min in PBS containing 300 mM NaCl, 1 mM MgCl₂ and 0.01% Tween-20. To remove unbound DNA, particles were washed three times with TE buffer supplemented with 0.01% Tween-20.

Next, streptavidin (IBA catalogue no. 2-0203-100) was loaded onto the particles at specific molar excesses in PBS buffer for 30 min (Supplementary Table 2). After washing to remove excess streptavidin, biotinylated EGFR (ACROBiosystems catalogue no. EGR-H82E3) was loaded in PBS for 30 min at room temperature (Supplementary Table 2). Particles were then washed two times with TE buffer containing 0.01% Tween-20 to remove unbound biotin-EGFR. The final particles were resuspended in 0.1 M TEAA containing 1% PVA for lyophilization. The lyophilized particles were stored at -80 °C until use. For tracking of surface protein after cell co-culture, streptavidin was further fluorescently labeled through a reaction with NHS-AFDye647 (Click Chemistry Tools, catalogue no. 16820) with an 8× molar excess at 37 °C for 1 h, followed by purification using size-exclusion chromatography (Thermo Scientific catalogue no. 89883).

Validation of surface functionalities

To detect EGFR antigens on the particle surface, particles were stained with a-EGFR antibody (Biolegend catalogue no. 352911 and no. 352917). The relative surface densities of AF647-labeled CD28 antibody and EGFR antigen were quantified by flow cytometry (Attune NxT cytometer). To detect the CAREp surface protein stability after T cell stimulation, co-cultured samples were collected within 7 days and imaged through confocal microscopy (Leica Stellaris 5 Confocal Microscope through 60× objectives). To determine the absolute antibody loading density, particles bearing CompR-antiCD28-AF647 via DNA-hybridization were dissociated in 7 M urea buffer at 70 °C for 30 min. Released protein, along with the standard, were denatured using the Laemmli buffer (Bio-Rad, catalogue no. 1610747) and detected using SDS-PAGE gel electrophoresis.

Generation of a-CD3 and a-CD28 coated PLGA microparticles and EGFR-coated magnetic particles

The functionalization of DNA-scaffolded microparticles with a-CD3 and a-CD28 antibodies were described previously^27,28. Briefly, anti-CD3 antibody (Bio X Cell, catalogue no. BE0001-2) was conjugated to a 3′-NH₂-modified complementary DNA strand (CompG, Supplementary Table 1) using the antibody-DNA conjugation strategy described above. CompG-a-CD3 and CompR-a-CD28 conjugates were subsequently hybridized onto DNA-scaffolded particles (G:R-[3:1] to generate a-CD3/a-CD28-coated PLGA microparticles. To generate EGFR-coated magnetic particles, biotinylated EGFR (ACROBiosystems, catalogue no. EGR-H82E3) was incubated with streptavidin-coated magnetic particles (ACROBiosystems, catalogue no. SMB-B01) at room temperature for 1 h, followed by three washes in PBS buffer.

Cell lines and culture

K562 (ATCC CCL-243), Nalm6 (ATCC CRL-3273), and U87 (ATCC HTB-14) cells were cultured in HyClone Dulbecco’s Modified Eagle Medium (Cytiva catalogue no. SH30243.01) with 10% FBS (Avantor catalogue no. 89510-186), 25 mM HEPES (Cytiva catalogue no. SH30237.01), and penicillin-streptomycin (Cytiva catalogue no. SV30010). These cell lines were further authenticated by Labcorp Inc. Lenti-X 293 T packaging cells were cultured in DMEM with 10%FBS. K562 cell line were transduced by lentivirus to stably express human truncated EGFR on surfaces and were FACS-sorted for the positive expression of the truncated EGFR transgenes. Nalm6 cell line were transduced by lentivirus to stably express human truncated EGFR on surfaces and luciferase reporter intracellularly, followed by FACS-sorting.

Primary human T cell isolation and culture

Peripheral blood was sourced from the Blood Centers of the Pacific. Primary CD4⁺ and CD8⁺ T cells were negatively selected from anonymized donor apheresis samples via commercial kits (STEMCELL Technologies; catalogue nos. 15062 and 15063). Human T cells are also sourced from the Human Immunology Core (HIC) at the University of Pennsylvania under Penn IRB protocol #705906. Human T cell engineering, stimulation, and characterization were carried out under Drexel University IRB protocol #21101. T cells were cryopreserved in 90% human AB serum (BIOIVT LLC catalogue no. MSPP-HP1022HI) and 10% DMSO. After thawing, T cells were cultured in human T-cell medium consisting of X-VIVO 15 (Lonza catalogue no. 04-418Q), 10 mM neutralized N-acetyl-l-cysteine (VWR catalogue no. BT124480-25), 5% human AB serum, and supplemented with 30 units/ml of IL-2 (PeproTech catalogue no. 200-02) for all experiments.

Transduction of human T cells with CAR constructs

Lentivirus was produced via transfection of Lenti-X 293 T cells with a transgene pHR’SIN: CSW expression vector and the viral packaging plasmids pCMVdR8.91 and pMD2.G using TransIT-VirusGEN Transfection Reagent (Mirus Bio catalogue no. MIR6700). Untransduced T cells were activated using CD3/CD28 Dynabeads (Life Technologies catalogue no. 11131D) at a 1:1 cell-to-bead ratio for 24 h prior to lentiviral transduction. On day 5 post-stimulation, Dynabeads were removed, and the T cells were expanded in culture until day 12 for cryopreservation. The EGFR-CAR construct is consisted of 7D12 nanobody ectodomain with a Myc-tag at the N-terminal, a CD8α transmembrane domain, and either 4-1BB or CD28 as the co-stimulatory domain, followed by a CD3ζ signaling domain. The BCMA-CAR construct is consisted of bb2121 scFv ectodomain with a Myc-tag at the N-terminal, a CD8α transmembrane domain, and a 4-1BB co-stimulatory domain, followed by a CD3ζ signaling domain. The CD19-CAR construct is consisted of FMC63 scFv ectodomain with a Myc-tag at the N-terminal, a CD8α transmembrane domain, and a 4-1BB co-stimulatory domain, followed by a CD3ζ signaling domain.

In vitro repeated stimulation of CAR-T cells

To evaluate the sustainability of CAR-T cell proliferation under different stimulation conditions, resting CAR-T cells were repeatedly stimulated with CAREp, CD3/CD28 Dynabeads, or tumor cells (U87 or K562-tEGFR). For each round of stimulation, cells at a density of 5 × 10⁵ cells per ml were seeded in a 24-well plate, with either CAREp at 2.5 particles per cell or tumor cells at a 1:1 cell-to-cell ratio. Cells were restimulated every 8–10 days upon resting from proliferation. Resting cells after varying rounds of stimulation were cryopreserved for subsequent analyses.

Antibody staining and flow cytometry analysis

All antibody staining for flow cytometry was performed in round-bottom 96-well tissue culture plates. Following centrifugation at 400 × g for 5 min to pellet the cells, the supernatant was carefully removed. Cells were then resuspended in 50 μl of PBS containing the desired fluorescently conjugated antibody and incubated for 25 min at 4 °C in the dark. After staining, cells were washed twice with PBS and resuspended in fresh PBS for acquisition on Attune NxT cytometer, BD FACSymphony A1, or Cytek Aurora flow cytometer. Data were analyzed with FlowJo v10.10.0. Gating strategies are provided in Supplementary Fig. 10.

For intracellular signaling analysis, CAR-T cells were collected after 0.5 and 24 h of stimulation with CAREp or U87 cells. Cells were immediately fixed by adding an equal volume of 8% paraformaldehyde (Electron Microscopy Sciences, catalogue no. 15710), followed by centrifugation and resuspension in 4% paraformaldehyde. Live/dead staining was performed 0.5 h prior to fixation (Invitrogen, catalogue no. 65-0865-14). After fixation, cells were permeabilized (BD, catalogue no. 555028) and stained for intracellular signaling markers. The vendors, catalogue number, clone name, and dilution of antibodies used in flow cytometry panels are listed in Supplementary Tables 3–5.

Bulk RNA sequencing analysis

EGFR CD28ζ-CAR-T cells were activated with either CAREp or U87 cells. Unstimulated CAR-T cells were used as a negative control. After 24 h, CAR-T cells were sorted out to remove leftover U87 cells, then RNA was extracted using RNeasy Plus Universal Mini Kit (Qiagen, catalogue no.73404). RNA libraries were prepared by Novogene (Durham, North Carolina). Prior to analysis, reads containing adapter sequences or more than 10% ambiguous (unidentifiable) bases were removed. Filtered reads were then aligned to the reference genome using HISAT2 (Hierarchical Indexing for Spliced Alignment of Transcripts, version 2). Gene expression levels were quantified as Fragments Per Kilobase of exon per Million mapped reads (FPKM) based on the aligned results. Additional downstream analyses—including Gene Ontology (GO) enrichment and differential gene expression analysis—were performed by Novogene to identify relevant biological pathways and gene sets. Gene Set Enrichment Analysis (GSEA) and Gene Set Variation Analysis (GSVA) were subsequently conducted using the normalized FPKM dataset provided.

Seahorse mitochondrial fitness assay

CAREp-stimulated CAR-T cells were thawed, resuspended in assay medium, and allowed to recover for 24 h prior to mitochondrial function analysis. The assay medium consisted of Seahorse XF RPMI medium (Agilent, catalogue no. 103576-100) supplemented with 2 mM L-glutamine (Agilent, catalogue no. 103579-100), 10 mM glucose (Agilent, catalogue no. 103577-100), and 1 mM sodium pyruvate (Agilent, catalogue no. 103578-100). Mitochondrial respiration and glycolytic activity were assessed by measuring oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) using the Seahorse XF96 Analyzer, following the manufacturer’s instructions. Briefly, 1.5 × 10⁵ cells in 100 μl assay medium were seeded in Seahorse XF96 cell culture microplates and incubated in a non-CO₂ incubator at 37 °C for 40 min prior to measurement. During the assay, cells were sequentially treated with oligomycin (1 μM), FCCP (1 μM), rotenone (0.5 μM), and antimycin A (0.5 μM), provided by the Seahorse XF Cell Mito Stress Test Kit (Agilent, catalogue no. 103010-100). Each cell condition was tested in six replicates.

In vitro cytotoxicity assay

CAR-T cell cytotoxicity was evaluated through co-cultures with K562-tEGFR or U87 tumor cells. CAR-T cells subjected to varying rounds of CAREp stimulation were co-cultured with target tumor cells in 96-well plates at effector-to-target (E:T) ratios of 1:1 or 1:4. U87 cell killing was assessed in real-time using the Incucyte S3 Live-Cell Analysis System (Sartorius), while K562-tEGFR cytotoxicity was quantified by flow cytometry using the Attune NxT cytometer at the end point. For flow cytometry analysis, constant volume was collected, and non-apoptotic cancer cells were gated by Myc-tag expression and Caspase-3/7 labeling (Invitrogen, catalogue no. C10432). Targeting killing efficacy is calculated by this formula:

$$\mathrm{Killing\; Rate}\left(\%\right)=\left(1-\frac{Tumorcellnumberintestedconditions}{TumorcellnumberinthecontrolgroupwithoutCAR-Tcell}\right)\times 100$$

In vivo tumor killing

All animal experiments were conducted in accordance with the guidelines and regulations of the Institutional Animal Care and Use Committee (IACUC) at Drexel University (protocol no. LA-24-130). All mice were housed in a specific pathogen-free (SPF) facility under a 12-hour light/12-hour dark cycle, with controlled ambient temperature (20 °C–22 °C) and humidity (30–70%). Animals from the experimental and control groups were co-housed throughout the study. To establish xenograft tumors, NSG mice (Jackson Laboratory, NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ stock no. 005557, ~6-8-week-old) were subcutaneously implanted with 1 × 10⁶ U87 cells on the right flank or intravenously implanted with 0.5 × 10⁶ Nalm6-EGFR cells. U87 tumor-bearing mice were randomly assigned to treatment and untreated control groups, with both sexes represented in each group (2–3 females and 2–3 males; n = 5 mice per group). For the Nalm6 model, female mice were used exclusively and were randomly assigned to treatment and control groups (n = 5 mice per group). To treat U87 tumor, 5 × 10⁵ CD4⁺ and 5 × 10⁵ CD8⁺ CAR-T cells were administered by intratumoral injection 14 days after tumor implantation. FACS-sorted unstimulated CAR-T cells (UnS) were frozen, thawed, and recovered 24 h prior to injection. Cells from CAREp-1 or Dynabeads-1 conditions were generated by stimulating frozen/thawed/recovered CAR-T cells (UnS) for 8 days prior to injection, and cells from CAREp-3 were generated by stimulating frozen/thawed/recovered CAREp-2 for 8 days prior to injection. U87 tumor growth was monitored by caliper measurements 2-3 times per week, and tumor volume was calculated using the formula: volume = length × width² / 2. To treat Nalm6-EGFR tumor, 0.6 × 10⁶ CD4⁺ and 0.6 × 10⁶ CD8⁺ CAR-T cells were injected intravenously at 4 days after tumor implantation. Tumor burden was monitored every 3-4 days by bioluminescence imaging (BLI) using the IVIS imaging system (Xenogen, catalogue no. 124262, with PerkinElmer Living Image Software v.4.5.4) after intraperitoneal injection of D-luciferin (GoldBio, catalogue no. LUCNA-1G). Mice body weight was measured once a week. Mice were euthanized before tumor volume approached the maximum allowable limit of 1500 mm³ as defined by institutional ethical guidelines, or upon reaching the predetermined experimental endpoint. For Nalm6 tumor-bearing mice, endpoints were defined as a loss of ≥20% of the initial body weight. Euthanasia was performed by CO₂ inhalation, followed by confirmation of respiratory arrest for a minimum of 3 min. Cervical dislocation was then performed as a secondary physical method to ensure death.

Telomere length and telomerase activity assays

CAR-T cells were outsourced to the UCSF core facility for the telomere length (T/S ratio) and telomerase activity (telomeric repeat amplification protocol, TRAP) analysis. For telomere length analysis, genomic DNA was extracted from frozen CAR-T cells, and quantitative PCR was performed to determine relative telomere length. This was calculated as the ratio of telomere repeat copy number to a single-copy gene (β-globin) copy number (denoted as “T/S”), which reflects the average telomere length relative to a reference DNA sample. As for the telomerase activity assay, the PCR conditions were as follows: 94 °C for 30 s, 60 °C for 30 s, and 72 °C for 45 s, repeated for 30 cycles. Amplified products were resolved on a 15% polyacrylamide gel in 0.5× TBE and visualized using SYBR Gold Nucleic Acid Gel Stain (Thermo Scientific, catalogue no. S11494). Relative telomerase activity is reported in arbitrary units (a.u.); Hela cell extracts were included as positive controls for inter-assay quality control and batch normalization when required.

Single-cell RNA sequencing and cell staining

Frozen cells from four timepoints of each donor (a total of 12 samples), including with and without varying rounds of CAREp stimulation, were thawed and stained by Human TruStain FcX (Biolegend, catalogue no. 422302). Next, CAR-T cells were individually stained by hashtag antibodies, and pooled together for staining with the TotalSeq-C Human Universal Cocktail (Biolegend, catalogue no. 399905). Beads barcoded single cells through microfluidic droplet formation then went through three types of library preparations, including whole transcriptome (GEX), CITE, and TCR, followed by pooled sequencing through one lane of NovaSeq6000 with a total of 3 billion reads. For data QC, barcode processing and unique molecular identifier (UMI) counting were performed using Cell Ranger version 8.0.1 (10X Genomics). Briefly, the ‘cellranger multi’ pipeline was run with three libraries: Gene Expression, VDJ-T, and Antibody Capture. References for Gene Expression and VDJ-T were Cellranger’s ‘refdata-gex-GRCh38-2020-A’ and ‘refdata-cellranger-vdj-GRCh38-alts-ensembl-8.0.1’, respectively, with a custom reference supplied for the Antibody Capture libraries. Gene expression and antibody capture outputs by ‘cellranger multi’ from multiple 10X lanes were concatenated in single matrices. Hashsolo was used to demultiplex the samples. The Scirpy package was used for clonotype analysis. Cells with mitochondrial RNA representing greater than 20%, as well as cells with less than 5% ribosomal RNA were excluded. The resulting dataset was normalized using a global scaling method converted by a scale factor (10,000) and log-transformed using the ‘log1p’ function in scanpy, and clustered using scanpy’s principal component and umap functions.

Single-cell RNA sequencing (scRNA-seq) data were analyzed using a combination of Python and R packages. Raw data generated from the Cell Ranger Multi pipeline were converted into both AnnData and 10x formats for downstream processing. Dimensionality reduction using Uniform Manifold Approximation and Projection (UMAP) and clustering using the Leiden algorithm were performed on the gene expression matrix. UMAP visualization and gene scoring were conducted primarily in Python using the Scanpy package. Subsequent analyses, including gene expression balloon plots, radar plots, violin plots, gene set variation analysis (GSVA), and bulk RNA-seq comparisons, were carried out in R using various packages such as Seurat, GSVA, and ComplexHeatmap, along with custom R functions.

Statistical analysis

Statistical analyses were performed using GraphPad Prism v10.4.2. Unless otherwise noted, all data are presented as mean ± standard deviation (s.d.). The following statistical tests were used where appropriate and specified in the figure legends: two-tailed paired ttest, two-tailed unpaired ttest, one-way ANOVA, and two-way ANOVA.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

X.H., T.A.D., Q.T., W.A.L., C.J.Y, and J.E. conceptualized and/or supervised the study. X.H. and Q.T. designed the experiments and interpreted the results. X.H., Q.Z., L.F., L.F., Y.C., H.H., J.Z.W., and R.H. performed the experiments including biomaterial engineering, T cell transduction, flow cytometry, cytotoxicity assay, and mitochondrial function analysis. L.F., T.M., and A.B. performed single-cell and bulk RNA sequencing analysis. X.H., Q.Z., S.M., and J.A.F designed and performed in vivo experiments testing CAR-T cell anti-tumor function in NSG mice. G.A., R.A., and R.H. contributed to experimental designs and CAR construct cloning. X.H., Q.Z., and J.A.F. wrote and edited the manuscript with input from all coauthors.

Peer review

Peer review information

Nature Communications thanks Emmanuel Donnadieu and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

We acknowledge the support from the National Institutes of Health Grant 1U54CA244438, Coulter-Drexel Translational Research Partnership Program, Margaret Q. Landenberger Research Foundation, and Drexel University Startup Fund.

Data availability

The Raw bulk and single-cell RNA sequencing datasets generated in this study have been deposited in the NCBI Sequence Read Archive (SRA) under BioProject accession code PRJNA1338258. The raw data for all figures (Source Data file 1), as well as the in vitro cell expansion profiles (Source Data file 2), are provided in the source data files. All data are included in the Supplementary Information or available from the authors, as are unique reagents used in this Article. The raw numbers for charts and graphs are available in the Source Data file whenever possible. Source data are provided with this paper.

Code availability

All custom code used to generate the heatmaps for the bulk RNA-seq and single-cell RNA-seq analyses has been deposited on GitHub and is publicly available at the following link: https://github.com/landon0-o/CAR-T-Cell-Data-Analysis.

Competing interests

X.H., T.A.D., Q.T., W.A.L., and J.Z.W. are inventors of pending patents related to the technology described in the manuscript. Q.Z., L.F., Y.C., T.M., H.H., G.A., R.A., J.Z.W., R.H., S.M., J.A.F., J.E., and C.J.Y. declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-76570-9.