Product Attributes of CAR T-cell Therapy Differentially Associate with Efficacy and Toxicity in Second-line Large B-cell Lymphoma (ZUMA-7) Open Access

Axicabtagene ciloleucel (axi-cel) is an autologous anti-CD19 chimeric antigen receptor (CAR) T-cell therapy approved for relapsed/refractory large B-cell lymphoma (LBCL; ref. 1) based on outcomes from ZUMA-1, the pivotal study of axi-cel in refractory LBCL after ≥2 lines of systemic therapy, and ZUMA-7 (NCT03391466; refs. 2–4), a phase III, randomized study investigating axi-cel versus standard of care (SOC; salvage chemotherapy and, in those with a response, high-dose chemotherapy and autologous stem-cell transplantation) in second-line relapsed/refractory LBCL. In ZUMA-7, axi-cel demonstrated superiority to SOC [event-free survival (EFS) HR, 0.398; P < 0.0001; median EFS, 8.3 months versus 2.0 months, respectively]. Axi-cel had manageable toxicity and rates of grade ≥3 cytokine release syndrome (CRS) and neurologic events (NE) were numerically lower than in ZUMA-1 (Cohorts 1+2; ref. 4). Despite overall positive outcomes from ZUMA-7, there remains a risk of primary resistance, relapse, and toxicities (CRS and NEs) following CAR T-cell intervention, warranting investigation of the product and patient characteristics in relationship to treatment resistance and toxicity.

In ZUMA-1 (third-line), CAR T-cell expansion [postinfusion peak and area under the curve from day 0 to 28 (AUC_0–28)] was associated with objective response rate (ORR) and durable responses, particularly when normalized to tumor burden (5). CAR T-cell expansion also associated with toxicities and concurrent elevated serum cytokines and chemokines (3). Furthermore, the number of infused product T cells with naive phenotype (CCR7⁺CD45RA⁺), normalized to pretreatment tumor burden associated with durable response, but not with high-grade CRS or NEs (5).

Here, we report pharmacokinetics, serum pharmacodynamics, and product (axi-cel product) and apheresis attributes associated with clinical outcomes in patients treated with axi-cel in ZUMA-7.

The ZUMA-7 (second-line) axi-cel pharmacokinetic profile was consistent with ZUMA-1 (third-line) as there was a direct association of CAR T-cell expansion with objective response (Fig. 1A; refs. 2, 3, 5). However, in ZUMA-7, duration of response (DOR) was not associated with CAR T-cell peak expansion (Supplementary Fig. S1). CAR T-cell peak levels in patients with ongoing response (n = 70) versus progression after response (n = 64) were also comparable (Fig. 1A) and higher than in patients with primary treatment resistance (n = 20). Similar results were observed for CAR T-cell AUC_0–28. In contrast to prior findings in third-line LBCL, CAR T-cell peak and/or AUC_0–28 normalized to tumor burden [per sum of product diameters (SPD)] did not correlate with durable response in second-line LBCL (Supplementary Fig. S2A). Although SPD was lower in ZUMA-7 (n = 150) versus ZUMA-1 (n = 101; ref. 6), CAR T-cell peak or AUC_0–28 normalized by SPD was comparable (Supplementary Fig. S2B). As later discussed, differences in association between pharmacokinetic peak or AUC_0–28 with outcome between ZUMA-1 and ZUMA-7 might be due to sampling differences. Notably, albeit with a reduced number of evaluable patients (n = 45) and with low detection levels, the number of CAR T cells/μL of blood on day 3 associated with ongoing response (Supplementary Fig. S3). Severity of toxicities by day 3 was not significantly associated with CAR T-cell expansion on day 3 (CRS, P = 0.48; NEs, P = 0.56).

CAR T-cell peak expansion significantly associated with grade ≥3 NEs (Fig. 1B), where median peak was 2.6-fold higher in patients with grade ≥3 NEs (n = 33) versus those with grade <3 NEs (n = 129). The rate of grade ≥3 NEs was comparable across patients in ongoing response [n = 16/75 (21.3%)], patients who progressed after response [relapsed; n = 15/66 (22.7%)], and nonresponders [n = 5/21 (23.8%)]. Median CAR T-cell peak was also 2-fold higher in a limited number of patients (n = 10) who experienced grade ≥3 CRS (Fig. 1C). This association between CAR T-cell peak expansion and grade ≥3 toxicities was consistent with ZUMA-1 (3).

Efficacy metrics, including any and ongoing response (Fig. 2A), EFS (Fig. 2B), progression-free survival (PFS) and DOR, were positively associated with naive T cells. The naive phenotype within axi-cel is referred to as being naive or naive-like, which, by established nomenclature, defines the lack of antigen exposure prior to interaction with antigen-presenting cells (7). However, the axi-cel product is polyclonally stimulated (CD3) during manufacturing, and it is anticipated that the naive phenotype corresponds to a more stem memory T-cell (T_SCM) phenotype (8) rather than a conventional antigen-naive phenotype. Flow cytometry analyses conducted herein cannot distinguish between naive T cells and T_SCM. Levels of naive T cells were elevated in patients with ongoing response (n = 72) versus patients who progressed after initial response (n = 65) and nonresponders (n = 21). Naive T cells significantly associated with improved EFS (Fig. 2A and B). Conversely, a product enriched (>median) with markers of late T-cell differentiation and exhaustion, like PD-1 and TIM-3 (9), corresponded with reduced EFS (Fig. 2C), PFS, and DOR. Notably, both CAR⁺ and CAR⁻ T cells expressing CCR7 and CD45RA, and the costimulatory molecules CD27 and CD28, were associated with improved efficacy (Fig. 2D; Supplementary Fig. S4), but not with high-grade CRS or NEs. Furthermore, the number of infused effector cells (T_EM⁺T_EFF, CCR7^–) associated with grade ≥3 NEs (n = 33; Fig. 2E). The number of central memory T cells (T_CM; CCR7⁺CD45RA^–) infused did not associate with grade ≥3 CRS (n = 10; Fig. 2F), which may be attributable, in part, to the limited number of patients that experienced high-grade CRS. Notably, when stratifying patients by grade 0–1 (n = 82) versus grade ≥2 CRS (n = 84), the number of central memory T cells infused was associated with grade ≥2 CRS (Supplementary Fig. S5).

Although the total number of CCR7⁺ cells infused moderately associated with peak CAR T-cell expansion (Spearman R ≈ 0.17; P < 0.05; Supplementary Fig. S6A), the association between CAR T-cell peak or AUC_0–28 and percentage of naive T cells in the product or total number of naive T cells infused did not reach significance by descriptive statistics (Spearman R ≈ 0.1; P ≈ 0.15). Notably, and given aforementioned data collection limitations, the number of CAR T cells/μL of blood on day 3 associated with the increased frequency of product naive T cells (Supplementary Fig. S6B).

In vitro cytotoxicity experiments were conducted to test whether a CAR T-cell product enriched with less-differentiated T cells could present improved antitumor activity. Healthy donor bulk CAR T-cell products were compared with donor-matched CAR T-cell products where CD45RO⁺ cells were removed using the bulk product as starting material. As expected, the enrichment process increased the proportion of naive T cells. The CAR transduction rate was comparable between enriched and bulk product (Supplementary Fig. S7A), as was T-cell proliferation upon coculture stimulation with CD19⁺ target cells (Supplementary Fig. S7B). Conversely, the naive-enriched product presented higher cytotoxicity in both short-term (Supplementary Fig. S7C) and long-term multiple stimulation (serial killing) assays (Supplementary Fig. S7D), and increased expression of costimulatory 4–1BB (Supplementary Fig. S7E) versus the bulk product.

A panel of 29 inflammatory and immune-modulatory analytes, preselected based on known CAR T-cell mechanisms of action and prior association with toxicity (5), was evaluated in serum. Analytes that were elevated in ≥50% of patients with ≥2-fold induction (peak) above baseline [the time point prior to optional bridging (limited to glucocorticoids only) and lymphodepleting chemotherapy] are depicted in Fig. 3A. The majority of analytes were rapidly and transiently elevated following axi-cel, reaching peak levels within 7 days and returning to baseline levels by week 4 postinfusion. Notably, IL15 and MCP-1 were elevated following lymphodepletion chemotherapy, prior to CAR T-cell infusion.

Peak and/or AUC_0–28 levels of several proinflammatory and immune-modulating analytes were associated with grade ≥3 CRS or NEs, including IL15, IL6, ferritin, GM-CSF, CXCL10, and IFNγ (Fig. 3B and C). Additional analytes that were elevated ≥2-fold (peak versus baseline) in a minority of patients associated with the occurrence of grade ≥3 toxicities, including ICAM-1 and VCAM-1. Overall, the ZUMA-7 pharmacodynamic profile was consistent with the known axi-cel mechanism of action and with ZUMA-1.

The association of peak or AUC_0–28 levels of analytes with efficacy (P < 0.05) was limited to fewer analytes than with toxicity (Fig. 3D and E). Postinfusion levels of IL10, IL5, and IL15 were positively associated with efficacy (ORR, CR, ongoing response, EFS, and PFS), whereas IL7, IL12p40, and MIP-1B were negatively associated with efficacy.

At baseline [the time point prior to optional bridging (limited to glucocorticoids only) and lymphodepleting chemotherapy] and on day 0 (the time point after lymphodepleting chemotherapy and prior to axi-cel infusion), elevated levels of TNFα, MIP-1A, granzyme B, MIP-1B, INTL2R/CD25, and IL7 correlated with reduced efficacy (CR, ongoing response, and ORR; Supplementary Fig. S8A). Notably, elevated levels of TNFα and INTL2R/CD25 also correlated with higher-grade toxicity (CRS and NE; Supplementary Fig. S8B).

At baseline (prior to optional bridging [limited to gluco­corticoids only] and lymphodepleting chemotherapy) and postinfusion time points, higher levels of inflammatory markers were associated (P < 0.05) with more differentiated product T cells (Supplementary Fig. S9A–S9B; Supplementary Fig. S10A). Specifically, higher levels of proinflammatory analytes associated with either percentage of product T_CM phenotype (CCR7⁺CD45RA^– T cells; Supplementary Fig. S9A) or T_EM+T_EFF phenotype (CCR7^– T cells; Supplementary Fig. S9B). Conversely, naive T-cell phenotype negatively associated with many inflammatory and immune-modulatory markers that were linked to toxicity (Supplementary Figs. S10B and S11).

Patients who experienced grade ≥3 NEs (n = 36) or CRS (n = 11) presented products that produced higher levels of IFNγ in coculture with CD19⁺ cells (normalized by transduction rate; Supplementary Fig. S12A and S12B). In contrast, lower IFNγ levels in coculture associated with increases in the naive product phenotype (Supplementary Fig. S12C). Although higher IFNγ coculture levels were enriched in patients exhibiting toxicity (Supplementary Fig. S12A and S12B and S13A and S13B), these levels did not correlate with best response of CR (Supplementary Fig. S13C), suggesting that product potency may more strongly underscore toxicity than efficacy.

T-cell phenotype in the apheresis and product were strongly correlated (Fig. 4A–D). Apheresis material enriched for CD27⁺CD28⁺CD8⁺ naive T cells further associated with objective response (P = 0.0369), and PFS (P = 0.0345), but not with CRS (P = 0.9619) or NEs (P = 0.1319; Fig. 4E–H). Similar trends were observed for ongoing response (P = 0.0545, Supplementary Fig. S14), EFS and DOR and were consistent in CD4⁺ T cells and CD3⁺ T cells, although significance was not reached.

CD27⁺CD28⁺ naive T-cell subpopulations from apheresis of ZUMA-7 (n = 137) were increased versus ZUMA-1 (n = 100), suggesting a higher probability of deriving products with a more stem-like phenotype in earlier lines of therapy (Supplementary Fig. S15A–S15C). Indeed, ZUMA-7 (n = 166) axi-cel products presented a higher percentage of naive T cells versus ZUMA-1 (n = 100; Supplementary Fig. S15D).

Here, we present associations among product features, pharmacokinetics, pharmacodynamics, and clinical outcomes from ZUMA-7, the first and largest randomized CAR T-cell trial in LBCL to date (2). An elevated percentage of naive T cells, or T_SCM (CCR7⁺CD45RA⁺) cells, in the product was associated with durable responses. Corroborating this result, patients in ZUMA-1 (third line) who had only 1–2 prior lines of therapy also had a higher percentage of naive T cells/T_SCM in both the product and apheresis versus patients who had 3+ prior lines of therapy (5). These findings compel earlier referral of patients with LBCL to CAR T-cell therapy, as this approach may avoid additional cycles of chemotherapy that could adversely impact CAR T-cell product fitness. T cells with a more naive T-cell/T_SCM phenotype may provide a better CAR T-cell therapeutic index; thus, next-generation products that enrich T_SCM are in development (10, 11). Considering the correlation of product naive-like T cells with efficacy and not with high-grade toxicities or inflammatory cytokines and chemokines, axi-cel in an earlier line of treatment may benefit patients by providing a starting material (apheresis) and subsequent product with higher fitness, including expression of CCR7, CD45RA, CD27, and CD28.

Axi-cel clinical efficacy outcomes reported through lines of therapy support this hypothesis (12). In ZUMA-12, patients with newly diagnosed high-risk LBCL and positive interim PET/CT scan were treated with axi-cel, demonstrating an 89% ORR and 78% CR rate with a manageable safety profile (12). In ZUMA-7 (second-line) and ZUMA-1 (third-line), axi-cel–treated patients had 83% ORR for both studies and CR rates of 65% and 58%, respectively (2, 4). The CR rate in ZUMA-7 was higher versus ZUMA-1, despite ZUMA-7 having a higher-risk patient population, with a high proportion with primary refractory disease (∼75%).

In product, a naive phenotype is associated with increased postinfusion CAR T-cell expansion, underscoring improved outcomes (5, 13). Conversely, although there was a modest association between total number of infused CCR7⁺ T cells (naive T cell + T_CM) and CAR T-cell peak or AUC_0–28, the frequency of naive T cells in the product did not significantly associate with CAR T-cell peak or AUC_0–28, nor with long-term persistence (≥3 months postinfusion). A limitation of the ZUMA-7 data versus ZUMA-1 (5) was the lack of consistent sampling around 2 weeks postinfusion (day 14), whereby only ∼15% (25/164) of evaluable patients had assessment at Week 2, potentially underestimating CAR T-cell expansion (2). Notably, in ZUMA-7, despite the findings being limited to 45 evaluable patients and low-level detection, an association between CAR T-cell levels in blood on day 3 and the percentage of product naive phenotype was observed. Day 3 CAR T-cell levels also significantly associated with ongoing response. Furthermore, in vitro postmanufacturing enrichment of naive cells (removal of CD45RO⁺ cells) consistently improved cytotoxicity, despite similar proliferation from nonenriched products, suggesting that these enriched products are more poised to generate the appropriate population of effector T cells that facilitate cytotoxicity with limited exhaustion. The translational value of in vitro experiments using healthy donor samples is limited and different results have been reported (14), which could be dependent on different methodologies. Nonetheless, these observations further indicate that improved tumor cell killing with a less differentiated CAR T-cell product may not entirely be dependent on the enhanced proliferative T-cell capacity.

Good and colleagues recently discussed that efficient CAR T-cell homing to the tumor is also critical and may result in lower CAR T-cell counts in blood, where patients with lower tumor burden can experience disease control with relatively low levels of peripheral CAR T-cell expansion (15). Indeed, a product enriched with naive T cells may provide enhanced lymph node/tumor homing (16). This is consistent with the recent observation that tumor T-cell infiltration impacts outcomes with CAR T-cell therapy (17).

In ZUMA-7, postinfusion CAR T-cell peak or AUC_0–28 associated with objective response and was comparable between patients in ongoing response and those who responded and then relapsed. CAR T-cell peak or AUC_0–28 did not associate with ongoing response even when normalized to tumor burden (per SPD). Recent data in third-line LBCL were consistent with ZUMA-7 where levels of postinfusion CAR T cells among patients who were in ongoing response were similar to those who relapsed after initial response to axi-cel (18). Notably, Good and colleagues did not observe any association between blood CAR T-cell levels and responses in third-line LBCL (15). Sampling, methodology, and disease context could explain the differences between studies, underlying the need for closer and consistent monitoring of CAR T-cell levels, particularly during the first few weeks postinfusion. Yet, although CAR T-cell expansion is important to trigger initial clinical response (consistent with ZUMA-1; ref. 3), the depth and duration of response is anticipated to rely on additional parameters, including tumor microenvironment features (18).

Previously, higher CAR T-cell expansion was suggested to help overcome high SPD in third-line LBCL to sustain durable responses (5, 19). SPD was reduced in second-line (ZUMA-7) versus third-line (ZUMA-1) LBCL (18); however, ZUMA-7 had a much higher proportion of patients with primary refractory disease. No association between SPD and efficacy was observed in second-line LBCL (ZUMA-7) among axi-cel–treated patients (6). This differs from the third-line setting (ZUMA-1), where axi-cel expansion correlated with ongoing response rate, particularly when normalized to SPD (5). This difference may be dependent on different product features and/or patient characteristics and tumor biology, including tumor burden and immune contexture, through lines of therapy, and between the two studies, as recently proposed (6). Notably, metabolic tumor volume (MTV) has been shown to negatively correlate with efficacy to axi-cel in both the third-line (20) and second-line (ZUMA-7) settings (though axi-cel remains superior to SOC for both high and low MTV; ref. 21), suggesting that MTV might represent a higher-resolution prognostic metric than SPD.

With a relatively high frequency of product naive T cells/T_SCM and other favorable features, including tumor microenvironment and reduced burden (SPD), durable responses may be less dependent on peripheral CAR T-cell quantity/expansion and more dependent on T-cell infiltration and sustained cytotoxicity within the tumor. This activity, particularly during the initial 2 to 4 weeks postinfusion, might be more relevant for a CAR T-cell product with a CD28 costimulatory domain, like axi-cel, versus other products, which may need longer-term persistence (22–24). A less differentiated product not only encompasses potentially higher proliferative capacity from the naive T-cell/T_SCM subtype (5, 13) but also presents costimulatory molecule enrichment and reduction of inhibitory checkpoint proteins/receptors, promoting in vivo differentiation and improved fitness across T-cell phenotypes.

In ZUMA-7, more-differentiated T cells (T_CM and T_EM+T_EFF) were associated with grade ≥2 CRS and Grade ≥3 NEs, respectively. Consistently, T_CM and T_EM+T_EFF in the product were associated with higher levels of proinflammatory and immune-modulatory serum molecules, including cytokines and chemokines. These data are consistent with others showing that enrichment of naive T cells in the CAR T-cell product can lead to improved efficacy and also to a more favorable toxicity profile with reduced cytokine release (10).

Collectively, these data and others support that naive T cells are a key population within CAR T-cell products that are more poised to expand, infiltrate lymphoid tumors, differentiate, and mediate sustained antitumor activity. This notion is consistent with prior observations that postinfusion CAR⁺ T cells with more prominent effector phenotypes are enriched in responding patients (25). However, the activity of different CAR T-cell products may rely on different mechanisms and, thus, our findings may not be generalizable to other CD19-tageting CAR T-cell therapies, such as lisocabtagene maraleucel, which has different manufacturing processes and costimulatory domains (26).

Baseline systemic inflammation was associated with reduced therapeutic index, where a number of elevated serum cytokines and chemokines negatively correlated with efficacy and positively correlated with high-grade CRS and NEs. Postinfusion elevated levels of a small number of serum analytes were associated with efficacy, compared with many analytes associated with high-grade toxicities. These observations support that the mechanisms leading to efficacy and toxicity of CAR T cells are not wholly overlapping and can be uncoupled. Potential targets for future interventional studies around toxicity management could include MCP-1, CXCL10, GM-CSF (27), IL17, IFNγ, TNFα, and/or sCD25/IL2R. Notably, preinfusion levels of IL2Ra/sCD25 and TNFα associated with both reduced efficacy and increased high-grade toxicity and were also negatively associated with product naive phenotype. Because baseline serum cytokines and chemokines were based on samples collected before optional bridging (limited to glucocorticoids only), conclusions regarding the effects of bridging on these analytes cannot be drawn from this analysis. Furthermore, relationships between tumor burden and baseline inflammatory markers have not been elucidated in this patient population and are being explored in ongoing analyses.

Although postinfusion elevated IL5, IL15, and IL10 were positively associated with efficacy, IL7, IL12, and MIP-1B were negatively associated with efficacy. Previously, patients with non-Hodgkin lymphoma with increased serum IL7 after CAR T-cell therapy were reported to have increased PFS, suggesting that IL7 signaling may enhance CAR T-cell durability (28). The opposite was observed in ZUMA-7, the reason for which is unclear but may include different products/constructs, patient characteristics, or number of prior lines of therapy.

MIP-1B is known for its role in myeloid cell activation and recruitment to inflamed tissues (29). Given the anticipated role of myeloid cells in mediating CAR T-cell toxicities (30–32), strategies to reduce MIP-1B release may help improve the CAR T-cell therapeutic index. Considering the canonical immune-suppressive role of IL10 (33), association between IL10 and CAR T-cell efficacy is controversial. Recent reports indicate IL10 may have antitumor activity, although the mechanisms are not fully elucidated and may be context-dependent (34–37). The direct link between lymphodepletion-dependent elevation in IL15 levels and improved response to CD19-directed CAR T-cell therapy was previously described (38, 39), where IL15 might preserve the CAR T-cell T_SCM phenotype and improve metabolic fitness (40). The possible association between IL5 and CAR T-cell efficacy is novel and warrants further investigation, and may represent accumulation of IL5 due to reduced sinking from B cells and granulocytes/myeloid cells in patients with improved outcomes (41).

In conclusion, we present evidence of an improved therapeutic index in patients with LBCL that have apheresis and CAR T-cell products enriched in T cells with a less-differentiated phenotype, namely expressing CCR7⁺, CD45RA⁺, and costimulatory receptors CD28 and CD27. This phenotype is more frequent in patients with lower systemic inflammation and in the setting of first- (12) and second-line therapy versus third or later lines, supporting the potential for improved outcomes with CAR T-cell therapy in earlier lines of therapy and the development of products with enrichment of naive T-cell/T_SCM phenotype. These findings provide insight and rationale for improved patient management and development of next-generation CAR T-cell therapeutics for patients with LBCL.

Evaluable samples from patients in the ZUMA-7 or ZUMA-1 Cohorts 1+2 safety analysis sets were analyzed. The ZUMA-7 safety analysis set was defined as randomized patients who received ≥1 dose of axi-cel (target dose, 2 × 10⁶ CAR T cells per kilogram of body weight or maximum of 2 × 10⁸ CAR positive viable T cells) or SOC. The ZUMA-1 safety analysis set was defined as all patients treated with any axi-cel dose. All patients provided written informed consent; the studies were approved by the institutional review board at each study site and were conducted according to Good Clinical Practice guidelines of the International Conference on Harmonization (2, 3).

ZUMA-7 efficacy (ORR, best response, EFS, PFS, DOR, ongoing response) and safety (CRS and NE rates) endpoints used the primary analysis data cutoff date, as previously described (2). EFS was defined as time from randomization to the earliest date of disease progression per Lugano Classification (42), commencement of new lymphoma therapy, or death from any cause. PFS was defined as time from randomization to disease progression or death from any cause. DOR was defined as time from first objective response to disease progression per Lugano Classification (42), commencement of new lymphoma therapy, or death from any cause. Ongoing response referred to patients who were in ongoing response by ZUMA-7 primary analysis data cutoff. Progression after response referred to patients who achieved a complete response or partial response and subsequently experienced disease progression. Nonresponders were defined as patients who achieved stable disease or progressive disease as best response, as previously reported (43).

As previously described (2), the severity of cytokine release syndrome was graded according to the modified criteria published by Lee and colleagues (44). Neurologic events were identified using prespecified search terms (45) and were graded per the Common Terminology Criteria for Adverse Events version 4.03.

To contextualize select findings, data from patients with evaluable samples in ZUMA-1 were included (follow-up 60 months).

Pharmacokinetic analysis of anti-CD19 CAR T cells (cells/μL blood) was performed by qPCR, as previously described (2, 5).

Product T-cell phenotypes and other attributes were measured in manufacturing as previously described (5). To strengthen the ZUMA-7 versus ZUMA-1 comparison of product T-cell phenotypes, ZUMA-1 samples were reanalyzed to ensure alignment with the ZUMA-7 gating strategy. Additional product characterization of costimulatory (CD27, CD28; gating strategy shown in Supplementary Fig. S16) and activation and exhaustion markers [programmed cell death protein-1 (PD-1), T-cell immunoglobulin and mucin domain-3 (TIM-3), lymphocyte activation gene-3 (LAG-3)] was performed by flow cytometry using a validated assay at CellCarta.

Apheresis T-cell phenotypes were calculated as a percentage of total T cells and identified by flow cytometry using a validated assay at CellCarta: central memory T cells (T_CM), naive T cells, effector memory T cells (T_EM), terminally differentiated effector memory (TEMRA) T cells, CD27⁺CD28⁺ naive CD4 T cells, CD27⁺CD28⁺ naive CD8 T cells, and CD27⁺CD28⁺ naive T cells.

Twenty-nine serum cytokines, chemokines, and other inflammatory markers were analyzed with validated Meso Scale Discovery methods (Medpace). Peak cytokine levels were defined as the maximum cytokine serum levels after baseline [the time point before optional bridging (limited to glucocorticoids only) and lymphodepleting chemotherapy] up to week 4 postinfusion. AUC of total cytokine levels from baseline to week 4 postinfusion was estimated using the trapezoidal rule.

Tumor burden was estimated by SPD of ≤6 target lesions per Cheson 2007 criteria and was assessed by central review (2, 46).

CAR T-cell products were generated from apheresis material from healthy donors (AllCells) following the axi-cel manufacturing protocol. Anti-CD19 CAR T effector cells derived from healthy donors were thawed and cultured overnight in RPMI-1640 (Corning) containing 10% FBS (Omega Scientific) before setting up coculture assays. Leukemia cell line Nalm-6 and diffuse large B-cell lymphoma cell line Toledo were obtained from ATCC. Cell authentication and Mycoplasma testing were performed regularly, during every cell thaw, using CellCheck (IDEXX BioAnalytics). The cell lines were transduced with vectors containing either luciferase gene or luciferase-GFP markers to obtain Nalm6-luc, Nalm6-luc-GFP and Toledo-LUC cell lines. All CD19⁺ target cells were cultured in RPMI1640 containing 10% FBS.

Naive T cells were isolated from healthy donor CAR T-cell products using the Naive Pan T Cell Isolation Kit (Miltenyi Biotec), an LS column and midiMACS separator according to manufacturer's instructions. The bulk product (unenriched) cells went through the same processing steps as the enriched CAR T-cell product, albeit in the absence of selecting antibodies. Cells were fluorescently stained using CD3-BUBV396, CCR7-PE, and CD45RA-BV510 and analyzed by flow cytometry using BD LSR Fortessa (BD Biosciences).

Purified T cells were labeled with 5 μL per 1 mL of cell suspension (10⁶/mL) with green CellBrite Cytoplasmic Membrane Dye for 20 minutes at 37°C. For analysis of T-cell proliferation, Toledo target cells were cocultured with T cells at varying effector-to-target ratios in 96-well flat-bottom plates at a final cell density of 10⁵ cells per well. A nuclear reactive dye Incucyte Nuclight Rapid NIR was added at 1:1,000 to the final cell culture, before being imaged every 2 hours for 6 days using an IncuCyte SX5 live cell analysis system (Essen Biosciences). T-cell proliferation was quantified as the change in nuclight staining at the end of the coculture compared with the beginning of the assay.

Cells were harvested 48 hours after coculture of CAR T-cell products and CD19⁺ target cells, Nalm-6, or Toldeo. Apoptosis and cell death were assayed using Annexin V-FITC and PI staining according to manufacturer's instructions (FITC Annexin V Apoptosis Detection Kit, BD Biosciences).

25,000 Nalm6-luc-GFP cells and CAR T-effector cells were cocultured based on effector-to-target ratios. Every 4 days at the end of each round, luminescence was measured after addition of 150 mg/mL luciferin (diluted 1:200) using luminometer and percentage of viable cells were quantified. At the end of each round, 25,000 target cells were added to all wells for later rounds. The process was continued for multiple rounds with serial stimulation of target cells until the effector cells are no longer effective.

Cells were collected 24 hours following coculture of CAR T cells with target cells. The cells were stained with following antibodies: CD3 (UCHT1)-BUV396, CD4 (SK3)-AF700, CD8 (SK1)-APCCy7, CCR7 (G043H7)-BV650, CD45RA (HI100)-BV510, Kip1 (CAR)-PE, 4–1BB (4B4–1-BV421), CD69 (H1.2F3)-BUV737, PD-1 (RMP1–30)-PECy7, and live/dead marker (FITC). For staining, antibodies were diluted according to manufacturer's instructions, added to the cells, and incubated for 30 minutes on ice. Data were acquired using BD LSR Fortessa (BD Biosciences).

All flow cytometry data analysis was performed using FlowJo 10 (Treestar). The plots were generated using GraphPad Prism 9.

Biomarkers from exploratory endpoints were analyzed for univariate associations between each covariate or with efficacy and safety endpoints. HR was calculated by the Cox regression model as a ratio of hazard of events between subgroups (≤median or Grade <3 as reference) or percentage increase in hazards for one-unit increase of continuous variables. OR was calculated by the logistic regression model as ratio of the odds of an outcome between subgroups (subgroups ≤median or grade <3 as reference) or as percentage increase in odds for one-unit increase of continuous variables. Kaplan–Meier plots and Cox regression were used to assess association between biomarkers and time-to-event endpoints. Wilcoxon rank sum test and logistic regression were used to assess association between biomarkers and binary endpoints. Kruskal–Wallis tests were used to assess association between biomarkers and categorical endpoints. Spearman rank-order correlation was used to assess association between two continuous biomarker covariates.

All P values are descriptive for post hoc analyses (P < 0.05 was considered significant). No adjustment for multiplicity testing was performed. For certain analyses, covariates were subdivided into subgroups by median value, as indicated herein.

Kite is committed to sharing clinical trial data with external medical experts and scientific researchers in the interest of advancing public health. As such, Kite shares anonymized individual patient data (IPD) upon request or as required by law or regulation. Such requests are at Kite's discretion and are dependent on the nature of the request, the merit of the research proposed, availability of the data, and the intended use of the data. If Kite agrees to the release of clinical data for research purposes, the requestor will be required to sign a data sharing agreement to ensure protection of patient confidentiality before the release of any data. For additional information or to make a request, contact [email protected].

S. Filosto reports as full time employee of Kite, a Gilead Company, and shareholder of Gilead Sciences. M.A. Canales reports personal fees from Beigene, BMS, Genmab, Kite, Incyte, Janssen, Lilly, Novartis, Sanofi, Roche, and personal fees from Takeda outside the submitted work. C.A. Portell reports grants from Kite, a Gilead Company, during the conduct of the study, and AbbVie, AstraZeneca, BeiGene, Kite/Gilead, and grants and personal fees from Merck outside the submitted work. C. To reports other support from Kite, a Gilead Company, and other support from Kite, a Gilead Company, during the conduct of the study and outside the submitted work. M. Schupp reports personal fees and other support from Gilead Sciences during the conduct of the study; and as employee of Kite, a Gilead Company, at the time of study and manuscript development. C.M. Warren reports other support from Kite, a Gilead Company, outside the submitted work. J. Budka reports personal fees from Kite, a Gilead Company, outside the submitted work. P. Cheng reports personal fees from Kite, a Gilead Company, outside the submitted work. J. Chou reports other support from Gilead and other support from Kyverna outside the submitted work. R.R. Shen reports other support from Kite, a Gilead Company, during the conduct of the study. J.R. Westin reports grants and personal fees from Kite/Gilead during the conduct of the study; grants and personal fees from ADC Therapeutics, AstraZeneca, and BMS, Morphosys/Incyte, Genentech, Novartis; personal fees from AbbVie, SeaGen, GenMab, Regeneron; and grants and personal fees from Nurix outside the submitted work. No disclosures were reported by the other authors.

S. Filosto: Conceptualization, formal analysis, validation, methodology, writing–original draft, writing–review and editing. S. Vardhanabhuti: Conceptualization, formal analysis, validation, methodology, writing–original draft, writing–review and editing. M.A. Canales: Resources, data curation, formal analysis, validation, writing-original draft, writing–review and editing. X. Poire: Resources, data curation, formal analysis, validation, writing–original draft, writing–review and editing. L.J. Lekakis: Resources, data curation, formal analysis, validation, writing–original draft, writing–review and editing. S. de Vos: Resources, data curation, formal analysis, validation, writing–original draft, writing–review and editing. C.A. Portell: Resources, data curation, formal analysis, validation, writing–original draft, writing–review and editing. Z. Wang: Formal analysis, validation, writing–original draft, writing–review and editing. C. To: Formal analysis, validation, writing–original draft, writing–review and editing. M. Schupp: Formal analysis, validation, writing–original draft, writing–review and editing. S. Poddar: Formal analysis, validation, writing–original draft, writing–review and editing. T. Trinh: Formal analysis, validation, writing–original draft, writing–review and editing. C.M. Warren: Formal analysis, validation, writing–original draft, writing–review and editing. E.G. Aguilar: Formal analysis, validation, writing–original draft, writing–review and editing. J. Budka: Formal analysis, validation, writing–original draft, writing–review and editing. P. Cheng: Formal analysis, validation, writing–original draft, writing–review and editing. J. Chou: Formal analysis, validation, writing–original draft, writing–review and editing. A. Bot: Conceptualization, formal analysis, validation, methodology, writing–original draft, writing–review and editing. R.R. Shen: Conceptualization, formal analysis, validation, methodology, writing–original draft, writing–review and editing. J.R. Westin: Conceptualization, resources, data curation, formal analysis, validation, methodology, writing–original draft, writing–review and editing.

We thank the patients who participated in this trial and their families, caregivers, and friends; the trial coordinators and health care staff at each site; Jennifer Yang, PhD, of Nexus Global Group Science, for medical writing assistance, funded by Kite, a Gilead Company; and all the employees of Kite, a Gilead Company, who were involved over the course of the trial for their contributions. This work was supported by Kite, a Gilead Company.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.