Association of baseline immune cell composition with CAR-T cell expansion and survival in Relapsed/Refractory large B-Cell lymphoma Authors: Xin-di Liu, He-nan Wang, Li-jie Zeng, Yuan-zheng Liang, Fu-li Li, Jin-yan Zhu, Jia Cong, Liang Wang Published: 2026 ABSTRACT BackgroundChimeric antigen receptor T (CAR-T) cell therapy has demonstrated remarkable efficacy in relapsed/refractory large B-cell lymphoma (R/R LBCL), yet nearly 40–60% of patients fail to achieve durable responses. The mechanisms underlying interpatient variability in CAR-T expansion and persistence remain incompletely understood. In this exploratory study, we preliminarily investigated the associations between baseline peripheral blood immune subsets, CAR-T expansion kinetics, and clinical outcomes.MethodsWe retrospectively analyzed 33 patients with R/R LBCL who received CD19/CD22 bispecific chimeric antigen receptor T-cell therapy (CAR2219) at our center. Peripheral blood samples were analyzed by flow cytometry. CAR-T cell expansion was monitored longitudinally, and group-based trajectory modeling (GBTM) classified patients into expansion patterns. Associations with progression-free survival (PFS) and overall survival (OS) were evaluated using Kaplan-Meier analysis and Cox regression. Twenty-two patients received tislelizumab (a PD-1 inhibitor) as maintenance therapy (200 mg intravenously every 3 weeks) starting on day 28 after CAR-T infusion. This regimen was not used as prior therapy, bridging therapy, or co-infusion, rather, it was designed to potentially enhance CAR-T persistence after the initial expansion phase. Given the small sample size (n = 33) and the exploratory nature of the analyses, all findings are hypothesis-generating only and require validation in large prospective cohorts.ResultsTwo distinct CAR-T expansion trajectories were identified: a low-expansion group (Group 1, n = 18) and a high-expansion group (Group 2, n = 15). Compared to Group 1, Group 2 exhibited higher peak expansion (Cmax, p < 0.001), greater total exposure (area under the curve (AUC), p < 0.001), and delayed time to peak (Tmax, 14 days vs. 10 days). Group 2 had higher baseline naive T cells (FDR-adjusted p = 0.024), helper T cells (FDR-adjusted p = 0.006), and CD4/CD8 ratio (FDR-adjusted p = 0.049), and fewer activated regulatory T cells (Tregs) (FDR-adjusted p = 0.018). Higher CD4/CD8 ratio associated with longer PFS (HR 0.41, 95% CI 0.17–0.73, p = 0.047). In exploratory subgroup analyses, a directional trend toward longer PFS was noted among patients with high baseline PD-1 expression who received PD-1 inhibitor maintenance therapy, whereas no such trend was observed in those without maintenance. These hypothesis-generating observations require validation in larger prospective cohorts.ConclusionBaseline immune cell composition may associate with CAR-T expansion and outcomes in R/R LBCL. Exploratory subgroup analyses suggested that the direction of association between baseline PD-1 expression and PFS may differ according to receipt of PD-1 inhibitor maintenance therapy (initiated on day 28 post-infusion), though no statistical significance was reached in either subgroup. Current evidence does not support the clinical use of baseline PD-1 expression as a predictive biomarker, and further validation in prospective studies is warranted.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12967-026-08394-6. Introduction Chimeric antigen receptor T-cell (CAR-T) therapy has changed the treatment landscape for patients with relapsed/refractory large B-cell lymphoma (R/R LBCL) who cannot have or have relapsed after autologous stem cell transplantation. Pivotal phase I and II clinical trials, including ZUMA-1 [1], JULIET [2], and TRANSCEND [3], have demonstrated that CAR-T cells can induce deep remissions in this heavily pretreated population. These landmark studies reported complete response (CR) rates ranging from 40% to 54%. Notably, a significant proportion of these responders achieve durable remissions, suggesting the potential for long-term disease control or even cure in a subset of patients. However, despite these remarkable advances, a substantial proportion of patients fail to achieve sustained responses, and the factors driving interpatient variability in CAR-T cell expansion, persistence, and clinical efficacy remain incompletely characterized. Emerging evidence indicates that the host immune microenvironment, particularly the composition of pre-apheresis peripheral blood, may significantly influence CAR-T cell product quality and subsequent in vivo expansion capabilities [4–6]. Single-cell profiling of CAR-T infusion products has further revealed that product composition, including T cell differentiation states and memory phenotypes, predicts relapse risk [7, 8]. The peripheral blood composition reflects the systemic immune status of the body, including the abundance of naive T cells, memory T cell subsets, and regulatory cell populations, which are factors that may affect the functional characteristics of the final CAR-T product [9, 10]. Patients with a more “robust” baseline immune status—characterized by a higher frequency of naive T cells and a favorable CD4/CD8 ratio—have been confirmed to be associated with superior CAR-T expansion and clinical outcomes [9]. Conversely, an elevated proportion of immunosuppressive regulatory T cells (Tregs) may weaken CAR-T expansion and promote tumor immune escape [10, 11]. In addition to T cell subset composition, expression of immune checkpoint molecules like programmed cell death protein 1 (PD-1) on baseline T cells may also affect CAR-T efficacy. Pre-existing T cell exhaustion, marked by high PD-1 expression, has been linked to inferior CAR-T expansion and persistence [12, 13]. However, the role of PD-1 expression in the context of concomitant PD-1 inhibitor therapy remains poorly defined, and whether baseline PD-1 status can guide combination strategies is an area of active investigation [14]. In this study, we gave the PD-1 inhibitor tislelizumab as maintenance therapy starting 28 days after CAR-T infusion – not as prior therapy, bridging, or co-infusion. This design was based on the following rationale. First, the early days after infusion are critical for CAR-T cell expansion and peak activation. Adding PD-1 blockade during this window may interfere with initial proliferation or exacerbate immune-related toxicities. Second, delayed PD-1 inhibition may help sustain CAR-T function after the contraction phase, potentially prolonging remission duration. This matches emerging clinical strategies that evaluating PD-1 blockade as a consolidation therapy following CAR-T cell therapy [15]. We aimed to comprehensively evaluate the relationship between pre-apheresis peripheral blood immune profiles, CAR-T expansion kinetics, and clinical outcomes in a cohort of patients with R/R LBCL receiving CD19/CD22 bispecific CAR-T therapy (CAR2219). Based on group-based trajectory modeling to objectively classify patients based on their CAR-T expansion patterns and examined associations with baseline immune subsets, progression-free survival (PFS), overall survival (OS), and treatment response. To further explore, we conducted subgroup analyses to preliminarily examine whether the prognostic significance of baseline PD-1 expression for PFS might be influenced by the addition of PD-1 inhibitor maintenance therapy. Methods Study design and patient population This retrospective study included 33 patients with R/R LBCL who received CAR2219 at our center between January 2025 and December 2025. Clinical follow-up was conducted until March 2026. Eligible patients were 14–85 years old and had relapsed or refractory large B-cell lymphoma according to the 2008 WHO Classification. Pathologically confirmed expression of CD19 and/or CD22 was required. Refractory disease was defined as progressive or stable disease as best response to the most recent chemotherapy regimen and relapsed disease was defined as the reappearance of radiographic evidence of disease after achieving complete remission. We excluded patients who had prior CAR-T or other gene therapy. Patients who received PD-1 inhibitor maintenance therapy after CAR-T treatment were included. Tislelizumab (200 mg intravenously every 3 weeks) was initiated on day 28 post-CAR-T infusion as maintenance therapy. The day 28 time point was selected to allow for the completion of the initial CAR-T expansion and activation phase (typically peaking around days 10–14), thereby avoiding potential interference with early CAR-T proliferation. Treatment was continued until disease progression, unacceptable toxicity (grade ≥ 3 immune-related adverse events), or patient withdrawal. PD-1 inhibitor was not administered as prior therapy, bridging therapy, or co-infusion in any patient. Assignment to PD-1 inhibitor maintenance therapy was non-randomized and based on institutional practice. Patients with high-risk features (≥3 prior lines or bulky disease) were more likely to receive combination therapy. This non-randomized assignment introduces potential selection bias, as patients with high-risk features were preferentially assigned to PD-1 inhibitor maintenance therapy. Given the limited sample size, the present analysis is exploratory, and all statistical results require confirmation in larger-scale cohorts. The study was conducted in accordance with the Declaration of Helsinki principles, and received approval from the Ethics Committee of Beijing Tongren Hospital, Capital Medical University (approval number: TREC2025–KY207). All patients provided written informed consent. Procedures All patients were treated with CAR2219, a third-generation bispecific CAR targeting CD19 and CD22. The construct has anti-CD19 and anti-CD22 single-chain variable fragments (scFvs) linked in tandem. The anti-CD19 scFv comes from the FMC63 clone. The anti-CD22 scFv also comes from the same clone but was modified for the tandem configuration. The anti-CD22 scFv was independently engineered and does not cross-react with CD19. Its functionality was validated using CD19-knockout target cells (Supplemental Methods). CAR expression and detection were performed via recognition of the CD19-specific FMC63 component. Downstream of the scFvs are a CD8α hinge and transmembrane domain, 4-1BB and CD3ζ cytoplasmic signaling domains, and a truncated epidermal growth factor receptor (tEGFR) reporter gene. The bispecific CAR is encoded by a self-inactivating lentiviral vector driven by an elongation factor 1 alpha (EF1-α) promoter (Supplementary Figure S1). Following leukapheresis, A 3-day lymphodepletion (cyclophosphamide: 250 mg/m2/d; fludarabine: 25 mg/m2/d) was followed by an intravenous dose of CAR2219 (2 × 106/kg). In addition to PD-1 inhibitor maintenance therapy, other concomitant medications administered during follow-up included exportin 1 (XPO1) inhibitors, Bruton’s tyrosine kinase (BTK) inhibitors, or histone deacetylase (HDAC) inhibitors, as per clinical practice. Clinical data collection Clinical and demographic data were extracted from electronic medical record system, including age, sex, Ann Arbor stage, B symptoms, International Prognostic Index (IPI) score, Eastern Cooperative Oncology Group performance status (ECOG PS), bulky disease (≥5 cm), number of prior lines of therapy, and history of PD-1 inhibitor administration. Best overall response was assessed per Lugano 2014 criteria and categorized as CR, partial response (PR), stable disease (SD), or progressive disease (PD). PFS was defined as the time from CAR-T infusion to disease progression or death from any cause, and OS was defined as the time from CAR-T infusion to death from any cause. Pre-apheresis immune cell subset analysis Peripheral blood samples were collected within 3 days prior to apheresis. Multicolor flow cytometry was performed using a panel of antibodies against surface markers to identify the following immune cell subsets: neutrophils, monocytes, lymphocytes, T lymphocytes (CD3+), B lymphocytes (CD3–CD19+), NK cells (CD3–CD56+), NKT lymphocytes (CD3+CD56+), naive T cells (CD3+CD45RA+CD62L+), helper T cells (CD3+CD4+), cytotoxic T cells (CD3+CD8+), central memory T cells (CD3+CD45RA-CD62L+) (Tcm), effector memory T cells (CD3+CD45RA-CD62L-) (Tem), activated T cells (CD3+HLA-DR+), Tregs (CD4+CD25+FoxP3+), activated regulatory T cells (CD4+CD25+FoxP3+CD45RA-), cytotoxic Tc cells (CD3+CD8+CD28+), and PD-1 expression levels. Naive T cells, Tcm, Tem and activated T cells are expressed as percentages of CD3+ T cells; Tregs and activated Tregs are expressed as percentages of CD4+ T cells; cytotoxic Tc cells are expressed as percentages of CD8+ T cells. The CD4/CD8 ratio is presented as a direct ratio. Flow cytometry was performed on a Beckman CytoFLEX analyzer, and absolute counts were determined using a two-platform method involving a hematology analyzer and flow cytometry. CAR-T cell expansion monitoring CAR-T cell expansion was longitudinally monitored by quantitative polymerase chain reaction (qPCR) for the CAR transgene and flow cytometry for CAR expression on peripheral blood T cells. Blood samples were collected at baseline (preinfusion), on days 3, 5, 7, 10, 14, 21, and 28 post-infusion, and monthly thereafter until CAR-T cells became undetectable or disease progression occurred. A nonlinear mixed-effects model was fitted to the CAR-T expansion data from each patient to calculate pharmacokinetic parameters, including maximum concentration (Cmax), time to maximum concentration (Tmax), and area under the curve (AUC). Statistical analysis All statistical analyses were performed using R software (version 4.1.2, R Foundation for Statistical Computing, Vienna, Austria). Graphical visualizations were generated primarily using the ggplot2, survminer, pheatmap, and forestplot packages. Based on CAR-T cell expansion kinetics, a group-based trajectory modeling (GBTM) approach was applied to classify patients into expansion subtypes. GBTM is a finite mixture model that identifies distinct longitudinal trajectories of a repeated measure (here, CAR-T cell concentration over time) by assuming the population has a finite number of latent groups, each with its own trajectory shape (linear, quadratic, or cubic). Models with 2, 3, and 4 groups were tested. For each candidate model, we specified a censored normal distribution for the repeated measures and a quadratic polynomial function to model the time-to-CAR-T relationship within each group. Model selection was guided by a combination of Bayesian Information Criterion (BIC), entropy (a measure of classification accuracy, with values > 0.7 indicating good separation), and average posterior probability of assignment (>0.7). The comparison of expansion parameters between groups was performed using the Wilcoxon rank-sum test, and AUC was estimated using the trapezoidal rule. Survival curves were estimated using the KaplanMeier method and compared between groups using the log-rank test. Comparisons of immune cell subsets between expansion groups (Group 1 vs. Group 2) were performed using the Wilcoxon rank-sum test. To account for multiple comparisons, the Benjamini-Hochberg false discovery rate (FDR) method was applied, and adjusted p values are reported. An FDR threshold of <0.05 was considered statistically significant. Univariate Cox proportional hazards regression models were used to assess the association between baseline immune subsets and PFS, with results presented as hazard ratios (HRs) and 95% confidence intervals (CIs). Due to the small sample size (n = 33) and low number of events, multivariable Cox regression incorporating multiple clinical variables was not feasible to avoid overfitting and unstable estimates. Key potential confounders including ECOG PS, number of prior lines of therapy, bulky disease, and lymphodepletion intensity were therefore not adjusted for in the main analysis. Instead, an exploratory limited multivariable Cox regression analysis adjusting only for the IPI score was conducted merely to illustrate directional trends, not to provide definitive confounding control. Exploratory subgroup analyses were performed stratified by receipt of PD-1 inhibitor maintenance therapy. All tests were two-sided, and a P value < 0.05 was considered statistically significant. Results Patient baseline characteristics A total of 33 patients with R/R LBCL were enrolled, and their baseline clinical and demographic characteristics are summarized in Table 1. The median age was 59 years (range, 16–81 years), with 17 males (51.5%) and 16 females (48.5%). The majority of patients presented with advanced-stage disease (Ann Arbor stage III–IV, 78.8%) and a high International Prognostic Index (IPI) score (≥3, 69.7%). Bulky disease (≥5 cm) was present in 27.3% of patients. The median number of prior lines of therapy was 2 (range, 1–5), with 24.2% of patients having received ≥ 3 prior lines. 66.7% of patients received PD-1 inhibitor maintenance therapy following CAR-T cell infusion. All patients underwent standard lymphodepleting chemotherapy followed by CAR2219. The median follow-up duration was 8.5 months (range, 1.0–13.1 months). The median PFS and OS were not reached. Meanwhile, Supplemental Table 1 shows that there were no significant differences between the low-expansion and high-expansion groups in key clinical variables, including age, sex, Ann Arbor stage, IPI score, ECOG PS, bulky disease, number of prior lines of therapy, and PD-1 inhibitor use (all p > 0.05).Table 1Baseline patient characteristicsTotal (n = 33)Age, median (range), y59 (16–81)Sex, n (%)Male17 (51.5)Female16 (48.5)Ann Arbor stage, n (%)I-II7 (21.2)III-IV26 (78.8)B symptoms, n (%)Present16 (48.5)Absent17 (51.5)IPI score, n (%)Low risk (0–2)10 (30.3)High risk (≥3)23 (69.7)ECOG PS, n (%)0–121 (63.6)≥212 (36.4)Bulky disease (≥5 cm), n (%)Yes9 (27.3)No24 (72.7)Prior lines of therapy, n (%)1–225 (75.8)≥38 (24.2)PD-1 inhibitor maintenance, n (%)Yes22 (66.7)No11 (33.3)Data are presented as n (%) unless otherwise specifiedOf the 22 patients receiving PD-1 inhibitor maintenance therapy, all initiated tislelizumab on day 28 post-infusion (200 mg intravenously every 3 weeks). None received PD-1 inhibitors prior to apheresis, during lymphodepletion, or concurrently with CAR-T infusionAbbreviations: IPI, International Prognostic Index; ECOG, Eastern Cooperative Oncology Group Table 1Baseline patient characteristicsTotal (n = 33)Age, median (range), y59 (16–81)Sex, n (%)Male17 (51.5)Female16 (48.5)Ann Arbor stage, n (%)I-II7 (21.2)III-IV26 (78.8)B symptoms, n (%)Present16 (48.5)Absent17 (51.5)IPI score, n (%)Low risk (0–2)10 (30.3)High risk (≥3)23 (69.7)ECOG PS, n (%)0–121 (63.6)≥212 (36.4)Bulky disease (≥5 cm), n (%)Yes9 (27.3)No24 (72.7)Prior lines of therapy, n (%)1–225 (75.8)≥38 (24.2)PD-1 inhibitor maintenance, n (%)Yes22 (66.7)No11 (33.3)Data are presented as n (%) unless otherwise specifiedOf the 22 patients receiving PD-1 inhibitor maintenance therapy, all initiated tislelizumab on day 28 post-infusion (200 mg intravenously every 3 weeks). None received PD-1 inhibitors prior to apheresis, during lymphodepletion, or concurrently with CAR-T infusionAbbreviations: IPI, International Prognostic Index; ECOG, Eastern Cooperative Oncology Group Baseline patient characteristics Data are presented as n (%) unless otherwise specified Of the 22 patients receiving PD-1 inhibitor maintenance therapy, all initiated tislelizumab on day 28 post-infusion (200 mg intravenously every 3 weeks). None received PD-1 inhibitors prior to apheresis, during lymphodepletion, or concurrently with CAR-T infusion Abbreviations: IPI, International Prognostic Index; ECOG, Eastern Cooperative Oncology Group Expansion trajectories correlate with survival outcomes Based on a group-based trajectory model, patients were stratified into two distinct CAR-T expansion patterns: a low-expansion group (Group 1, n = 18) and a high-expansion group (Group 2, n = 15). Longitudinal CAR-T expansion curves for all patients, stratified by trajectory group, are shown in Fig. 1A. Patients in Group 2 exhibited a significantly higher Cmax compared with those in Group 1 (P < 0.001). The total exposure, as measured by AUC, was also significantly greater in Group 2 than in Group 1 (P < 0.001). In addition, Tmax was delayed in Group 2 patients (P = 0.008). The interquartile range (IQR) of expansion values was wider in Group 2 patients (Table 2). The 2-group trajectory model was selected after testing 2-group, 3-group, and 4-group solutions. The 3-group and 4-group models failed to converge stably due to the limited sample size (n = 33). The 2-group solution demonstrated excellent classification accuracy, with an entropy of 0.82 and average posterior probabilities of 0.91 for Group 1 and 0.88 for Group 2, all exceeding the conventional threshold of 0.7. The two groups were well-balanced (n = 18 and n = 15). Therefore, the 2-group solution was selected for subsequent analyses. Kaplan-Meier analysis showed that patients in Group 2 had a numerically longer but statistically non-significant PFS compared with Group 1 (P = 0.053, log-rank test underpowered due to sample size; Fig. 1B). Similarly, OS was more favorable in Group 2 than in Group 1, although the difference did not reach statistical significance (P = 0.2, log-rank test underpowered due to sample size; Fig. 1C). These differences did not reach statistical significance and should not be interpreted as evidence of a true difference.Fig. 1CAR-T expansion kinetics and clinical outcomes according to expansion trajectory groups. (A) longitudinal CAR-T expansion curves for all patients, stratified by trajectory group. The y-axis represents CAR-T cell concentration in peripheral blood (cells/L) on a logarithmic scale. Solid lines represent fitted values from nonlinear mixed-effects models, and shaded areas represent 95% confidence intervals. Patients were classified into two distinct expansion patterns using group-based trajectory modeling (GBTM), a finite mixture modeling approach that identifies latent subgroups with similar longitudinal patterns. The 2-group solution was selected based on convergence, entropy (0.82), and average posterior probabilities (>0.87 for both groups): Group 1 (low expansion, n = 18, red) and Group 2 (high expansion, n = 15, blue). The bar charts below show total exposure (area under the curve), peak expansion, and time to peak for the two groups. (B) progression-free survival according to expansion group. Kaplan-Meier curves showing progression-free survival (PFS) for patients in Group 1 (low expansion) and Group 2 (high expansion). The Kaplan-Meier estimates become less stable beyond day 300 due to the limited number of patients remaining at risk. (C) overall survival according to expansion group. Kaplan-Meier curves showing overall survival (OS) for patients in Group 1 and Group 2Table 2CAR-T expansion parameters stratified by trajectory groupParameterGroup 1(Low expansion, n = 18)Group 2(High expansion, n = 15)Fold change / DifferenceP valueAUC (×108 cells/L × days), median (IQR)30.23 (8.23–58.92)85.47 (40.76–421.68)~2.83-fold ↑<0.001Cmax (×108 cells/L), median (IQR)2.62 (0.43–8.77)7.46 (2.92–30.12)~2.85-fold ↑<0.001Tmax (days), median (IQR)10 (7–14)14 (9–18)+4 days0.008IQR rangeNarrowerWider--Data are presented as median (interquartile range [IQR]) unless otherwise specifiedFold change was calculated as the ratio of medians between Group 2 and Group 1; difference in Tmax is presented as absolute difference in mediansAbbreviations: AUC, Area under the curve; Cmax, Maximum concentration; Tmax, Time to maximum concentration; IQR, Interquartile range CAR-T expansion kinetics and clinical outcomes according to expansion trajectory groups. (A) longitudinal CAR-T expansion curves for all patients, stratified by trajectory group. The y-axis represents CAR-T cell concentration in peripheral blood (cells/L) on a logarithmic scale. Solid lines represent fitted values from nonlinear mixed-effects models, and shaded areas represent 95% confidence intervals. Patients were classified into two distinct expansion patterns using group-based trajectory modeling (GBTM), a finite mixture modeling approach that identifies latent subgroups with similar longitudinal patterns. The 2-group solution was selected based on convergence, entropy (0.82), and average posterior probabilities (>0.87 for both groups): Group 1 (low expansion, n = 18, red) and Group 2 (high expansion, n = 15, blue). The bar charts below show total exposure (area under the curve), peak expansion, and time to peak for the two groups. (B) progression-free survival according to expansion group. Kaplan-Meier curves showing progression-free survival (PFS) for patients in Group 1 (low expansion) and Group 2 (high expansion). The Kaplan-Meier estimates become less stable beyond day 300 due to the limited number of patients remaining at risk. (C) overall survival according to expansion group. Kaplan-Meier curves showing overall survival (OS) for patients in Group 1 and Group 2 Table 2CAR-T expansion parameters stratified by trajectory groupParameterGroup 1(Low expansion, n = 18)Group 2(High expansion, n = 15)Fold change / DifferenceP valueAUC (×108 cells/L × days), median (IQR)30.23 (8.23–58.92)85.47 (40.76–421.68)~2.83-fold ↑<0.001Cmax (×108 cells/L), median (IQR)2.62 (0.43–8.77)7.46 (2.92–30.12)~2.85-fold ↑<0.001Tmax (days), median (IQR)10 (7–14)14 (9–18)+4 days0.008IQR rangeNarrowerWider--Data are presented as median (interquartile range [IQR]) unless otherwise specifiedFold change was calculated as the ratio of medians between Group 2 and Group 1; difference in Tmax is presented as absolute difference in mediansAbbreviations: AUC, Area under the curve; Cmax, Maximum concentration; Tmax, Time to maximum concentration; IQR, Interquartile range CAR-T expansion parameters stratified by trajectory group Data are presented as median (interquartile range [IQR]) unless otherwise specified Fold change was calculated as the ratio of medians between Group 2 and Group 1; difference in Tmax is presented as absolute difference in medians Abbreviations: AUC, Area under the curve; Cmax, Maximum concentration; Tmax, Time to maximum concentration; IQR, Interquartile range Comparison of immune cell subsets between expansion groups We characterized the baseline peripheral blood immune profile of the entire cohort. Figure 2A presents the mean values for each immune subset across all 33 patients. Of note, five subsets (neutrophils, T lymphocytes, B lymphocytes, helper T cells, and cytotoxic T cells) are reported as absolute counts (×109 cells/L), while all other subsets are reported as percentages with denominators as specified in the Methods section. The mean absolute counts were as follows: neutrophils: 3.00 × 109 cells/L; T lymphocytes: 0.63 × 109 cells/L; B lymphocytes: 0.01 × 109 cells/L; helper T cells: 0.32 × 109 cells/L; cytotoxic T cells: 0.31 × 109 cells/L. The extremely low mean absolute B lymphocyte count (0.01 × 109cells/L) reflects the underlying B-cell malignancy and the suppressive effects of prior lines of therapy. For subsets reported as percentages, the mean values were as follows: CD4/CD8 ratio: 0.75; naive T cells: 18.86% of CD3+ T cells; Tcm: 28.67% of CD3+ T cells; Tem: 32.34% of CD3+ T cells; activated T cells: 45.7% of CD3+ T cells; Tregs: 58.7% of CD4+ T cells; activated Tregs: 39.96% of CD4+ T cells; PD-1 expression: 28.97%; cytotoxic Tc cells: 39.45% of CD8+ T cells. These baseline values reflect the expected immune alterations in heavily pretreated R/R LBCL patients, including severe B-cell lymphopenia, moderate T-cell lymphopenia, and enrichment of effector memory and activated T cell subsets.Fig. 2Baseline peripheral blood immune cell profiles according to CAR-T expansion trajectory groups. (A) distribution of immune cell subsets. Bar charts showing the distribution of immune cell subsets in peripheral blood prior to apheresis. Bars represent mean values per group, with mean values annotated above each bar. Naive T cells, Tcm, Tem and activated T cells are expressed as percentages of CD3+ T cells; Tregs and activated Tregs are expressed as percentages of CD4+ T cells; cytotoxic Tc cells are expressed as percentages of CD8+ T cells. The CD4/CD8 ratio is presented as a direct ratio. (B) comparison of immune cell subsets between expansion groups. Box plots showing the distribution of immune cell subsets in Group 1 (low expansion, n = 18) and Group 2 (high expansion, n = 15). Points represent individual observations, and bars represent mean values per group, with mean values (rounded to two decimal places) annotated above each bar. Raw p values (Wilcoxon rank-sum test) and FDR-adjusted p values are displayed. (C) correlation between immune cell subsets and clinical outcomes. Spearman correlation heatmap showing associations between baseline immune cell subsets and key clinical endpoints, including trajectory group classification, best overall response, progression-free survival (PFS), and overall survival (OS). Red indicates positive correlation, blue indicates negative correlation. Correlations with p < 0.05 are outlined in black. Only immune subsets with statistically significant correlations (P < 0.05) with at least one clinical endpoint are shown. Non-significant subsets have been omitted to improve readability and highlight the most relevant associations Baseline peripheral blood immune cell profiles according to CAR-T expansion trajectory groups. (A) distribution of immune cell subsets. Bar charts showing the distribution of immune cell subsets in peripheral blood prior to apheresis. Bars represent mean values per group, with mean values annotated above each bar. Naive T cells, Tcm, Tem and activated T cells are expressed as percentages of CD3+ T cells; Tregs and activated Tregs are expressed as percentages of CD4+ T cells; cytotoxic Tc cells are expressed as percentages of CD8+ T cells. The CD4/CD8 ratio is presented as a direct ratio. (B) comparison of immune cell subsets between expansion groups. Box plots showing the distribution of immune cell subsets in Group 1 (low expansion, n = 18) and Group 2 (high expansion, n = 15). Points represent individual observations, and bars represent mean values per group, with mean values (rounded to two decimal places) annotated above each bar. Raw p values (Wilcoxon rank-sum test) and FDR-adjusted p values are displayed. (C) correlation between immune cell subsets and clinical outcomes. Spearman correlation heatmap showing associations between baseline immune cell subsets and key clinical endpoints, including trajectory group classification, best overall response, progression-free survival (PFS), and overall survival (OS). Red indicates positive correlation, blue indicates negative correlation. Correlations with p < 0.05 are outlined in black. Only immune subsets with statistically significant correlations (P < 0.05) with at least one clinical endpoint are shown. Non-significant subsets have been omitted to improve readability and highlight the most relevant associations Patients in Group 2 exhibited significantly higher frequencies of naive T cells (CD3+CD45RA+CD62L+; p = 0.008, FDR-adjusted p = 0.024), helper T cells (CD3+CD4+; p = 0.001, FDR-adjusted p = 0.006), and CD4/CD8 ratios (P = 0.037, FDR-adjusted p = 0.049) compared with those in Group 1. Conversely, the frequency of activated Tregs was significantly lower in Group 2 (CD4+CD25+FoxP3+CD45RA-; p = 0.002, FDR-adjusted p = 0.018). Tcm trended higher in Group 2 (mean 28.92% vs. 25.74%), while Tem were comparable between groups (33.44% vs. 33.55%). PD-1 expression did not differ significantly between groups (28.49% vs. 30.63%, p = 0.762). (Fig. 2B). Correlation and Cox regression Analyses To systematically evaluate the relationships between baseline immune parameters and clinical endpoints, we performed Spearman correlation analysis. Figure 2C presents the correlation heatmap, which includes only immune subsets with at least one statistically significant correlation (P < 0.05) with the four clinical endpoints: trajectory group classification (Group 2 vs. Group 1), best response (ordinal: CR > PR >SD >PD), PFS (days), and OS (days). CD4/CD8 ratio was positively correlated with group classification (ρ = 0.36, p < 0.5), PFS (ρ = 0.03, p < 0.5) and OS (ρ = 0.02, p < 0.5). Naive T cells demonstrated similar positive correlations with group classification (ρ = 0.46, p < 0.1) and PFS (ρ = 0.09, p < 0.5). Helper T cells showed moderate positive correlations with group classification (ρ = 0.58, p < 0.001), PFS (ρ = 0.09, p < 0.5) and OS (ρ = 0.19, p < 0.5). Activated Tregs showed moderate negative correlations with group classification (ρ = −0.54, p < 0.1) and PFS (ρ = −0.06, p < 0.5), consistent with their immunosuppressive function (Fig. 2C). Univariate Cox regression analysis (Table 3) demonstrated that a higher CD4/CD8 ratio was significantly associated with a reduced risk of progression or death (HR, 0.41; 95% CI, 0.17–0.73; p = 0.047). The frequency of helper T cells showed a trend toward a protective effect (HR, 0.67; 95% CI, 0.45–0.99; p = 0.054), as did naive T cells (HR, 0.78; 95% CI, 0.60–1.01; p = 0.060). Activated regulatory T cells showed a trend toward increased risk (HR, 1.32; 95% CI, 1.00–1.74; p = 0.057). Other immune subsets, including B lymphocytes, cytotoxic T cells, Tcm, Tem, and PD-1 expression, did not show statistically significant associations in the univariate analysis. After adjustment for the IPI score, the CD4/CD8 ratio remained associated with PFS (HR 0.44, 95% CI 0.19–0.79, p = 0.047); helper T cells maintained a protective trend (HR 0.69, 95% CI 0.51–0.97, p = 0.054). The directional trends for naive T cells and activated Tregs remained consistent with those observed in univariate analysis. Given the extremely limited sample size and low number of events, a limited multivariable analysis adjusting only for IPI score was performed as an exploratory attempt to illustrate directional trends (Supplemental Table 2).Table 3Univariate Cox regression analysis for progression-free survivalVariableUnivariate AnalysisHR (95% CI)P valueImmune cell subsetsNeutrophils1.81 (0.37–8.89)0.470T lymphocytes0.83 (0.29–2.37)0.408B lymphocytes1.59 (0.32–7.81)0.532Helper T cells0.67 (0.45–0.99)0.054Cytotoxic T cells1.92 (0.87–4.22)0.156Naive T cells0.78 (0.60–1.01)0.060Tcm0.95 (0.25–3.61)0.500Tem0.97 (0.19–4.98)0.863Activated T cells1.00 (0.12–8.33)0.942Tregs1.04 (0.99–1.10)0.170Activated Tregs1.32 (1.00–1.74)0.057Tc0.97 (0.95–1.00)0.188PD-11.01 (0.21–4.85)0.569CD4/CD8 ratio0.41 (0.17–0.73)0.047Due to the limited sample size (n = 33), multivariate analysis was not performed to avoid overfittingP values < 0.05 were considered statistically significant; P values between 0.05 and 0.10 are reported as trends toward associationAbbreviations: CI, confidence interval; HR, hazard ratio; Tcm, central memory T cells; Tem, effector memory T cells; Tregs, regulatory T cells; Tc, Cytotoxic Tc cells Table 3Univariate Cox regression analysis for progression-free survivalVariableUnivariate AnalysisHR (95% CI)P valueImmune cell subsetsNeutrophils1.81 (0.37–8.89)0.470T lymphocytes0.83 (0.29–2.37)0.408B lymphocytes1.59 (0.32–7.81)0.532Helper T cells0.67 (0.45–0.99)0.054Cytotoxic T cells1.92 (0.87–4.22)0.156Naive T cells0.78 (0.60–1.01)0.060Tcm0.95 (0.25–3.61)0.500Tem0.97 (0.19–4.98)0.863Activated T cells1.00 (0.12–8.33)0.942Tregs1.04 (0.99–1.10)0.170Activated Tregs1.32 (1.00–1.74)0.057Tc0.97 (0.95–1.00)0.188PD-11.01 (0.21–4.85)0.569CD4/CD8 ratio0.41 (0.17–0.73)0.047Due to the limited sample size (n = 33), multivariate analysis was not performed to avoid overfittingP values < 0.05 were considered statistically significant; P values between 0.05 and 0.10 are reported as trends toward associationAbbreviations: CI, confidence interval; HR, hazard ratio; Tcm, central memory T cells; Tem, effector memory T cells; Tregs, regulatory T cells; Tc, Cytotoxic Tc cells Univariate Cox regression analysis for progression-free survival Due to the limited sample size (n = 33), multivariate analysis was not performed to avoid overfitting P values < 0.05 were considered statistically significant; P values between 0.05 and 0.10 are reported as trends toward association Abbreviations: CI, confidence interval; HR, hazard ratio; Tcm, central memory T cells; Tem, effector memory T cells; Tregs, regulatory T cells; Tc, Cytotoxic Tc cells PD-1 expression and interaction with PD-1 inhibitor therapy Figure 3A–B shows exploratory subgroup analyses by PD-1 inhibitor maintenance therapy. In patients without maintenance (n = 11), those with PFS < 246 days had higher baseline PD-1 expression than those with PFS ≥ 246 days (median 28.6% vs. 19.1%; p = 0.19). In patients who got maintenance (n = 22), the pattern reversed (26.4% vs. 36.6%; p = 0.13).Fig. 3Exploratory analysis of baseline PD-1 expression and progression-free survival (PFS), stratified by PD-1 inhibitor maintenance therapy. (A) box plots showing baseline PD-1 expression (%) in patients stratified by PD-1 inhibitor maintenance therapy (yes/no) and PFS duration (<246 days vs. ≥246 days). Among patients who did not receive PD-1 inhibitor maintenance therapy (n = 11), those with shorter PFS had numerically higher baseline PD-1 expression (median 8.6% vs. 19.1%; p = 0.19). Among patients who received PD-1 inhibitor maintenance therapy (n = 22), baseline PD-1 expression was similar between PFS subgroups (median 26.4% vs. 36.6%; p = 0.13). p values were calculated using the Wilcoxon rank-sum test. (B) Kaplan-Meier curves of PFS comparing high vs. low baseline PD-1 expression, stratified by PD-1 inhibitor maintenance therapy. High and low PD-1 expression were defined using the median baseline PD-1 expression of the entire cohort (28.97%). Left panel: patients who did not receive PD-1 inhibitor maintenance therapy (n = 11). No statistically significant difference was observed between high and low PD-1 expression groups (P = 0.55). Right panel: patients who did receive PD-1 inhibitor maintenance therapy (n = 22). A numerical trend toward longer PFS was observed in patients with high baseline PD-1 expression compared to those with low expression (P = 0.24), with the high PD-1 group showing a consistently higher survival curve. None of these differences reached statistical significance, and these exploratory findings should be interpreted with caution due to the limited sample size. Shaded areas represent 95% confidence intervals Exploratory analysis of baseline PD-1 expression and progression-free survival (PFS), stratified by PD-1 inhibitor maintenance therapy. (A) box plots showing baseline PD-1 expression (%) in patients stratified by PD-1 inhibitor maintenance therapy (yes/no) and PFS duration (<246 days vs. ≥246 days). Among patients who did not receive PD-1 inhibitor maintenance therapy (n = 11), those with shorter PFS had numerically higher baseline PD-1 expression (median 8.6% vs. 19.1%; p = 0.19). Among patients who received PD-1 inhibitor maintenance therapy (n = 22), baseline PD-1 expression was similar between PFS subgroups (median 26.4% vs. 36.6%; p = 0.13). p values were calculated using the Wilcoxon rank-sum test. (B) Kaplan-Meier curves of PFS comparing high vs. low baseline PD-1 expression, stratified by PD-1 inhibitor maintenance therapy. High and low PD-1 expression were defined using the median baseline PD-1 expression of the entire cohort (28.97%). Left panel: patients who did not receive PD-1 inhibitor maintenance therapy (n = 11). No statistically significant difference was observed between high and low PD-1 expression groups (P = 0.55). Right panel: patients who did receive PD-1 inhibitor maintenance therapy (n = 22). A numerical trend toward longer PFS was observed in patients with high baseline PD-1 expression compared to those with low expression (P = 0.24), with the high PD-1 group showing a consistently higher survival curve. None of these differences reached statistical significance, and these exploratory findings should be interpreted with caution due to the limited sample size. Shaded areas represent 95% confidence intervals Kaplan-Meier analysis (Fig. 3B) showed no statistically significant association between baseline PD-1 expression and PFS in either subgroup, most likely due to the small sample size. Among patients without maintenance (n = 11), there was no clear difference in PFS between high and low baseline PD-1 groups (log-rank p = 0.55). Among patients who got maintenance (n = 22), a numerical trend toward longer PFS was observed in patients with high baseline PD-1 expression compared to those with low expression (P = 0.24), with the high PD-1 group showing a consistently higher survival curve. However, these differences were not statistically significant, and no definitive conclusions can be drawn given the exploratory nature of this analysis and the small subgroup sample sizes. These hypothesis-generating observations warrant validation in larger prospective cohorts. Discussion In this exploratory study, we observed preliminary associations between pre-apheresis peripheral blood immune profiles and CAR-T cell expansion kinetics and clinical outcomes in patients with R/R LBCL, although these findings require confirmation in larger cohorts. Using GBTM, we identified two distinct CAR-T expansion patterns. Tmax was delayed in Group 2 patients, suggesting more sustained expansion, a pattern consistent with the expansion kinetics observed when CAR-T products contain a higher proportion of naive/memory T cells [16–18]. These expansion patterns were associated with baseline immune subsets, including higher frequencies of naive T cells, helper T cells, and CD4/CD8 ratio, as well as lower frequencies of activated Tregs. Furthermore, exploratory subgroup analyses suggested a potential association between baseline PD-1 expression and maintenance therapy with PD-1 inhibitors. The association between baseline T-cell composition and CAR-T expansion is biologically plausible. CAR-T products are derived from autologous T cells obtained via apheresis, and the quality of the starting cell population directly determines the functional characteristics and therapeutic potential of the final product [19, 20]. Previous studies have shown that a higher proportion of naive T cells in the apheresis product is associated with improved CAR-T cell proliferative capacity and persistence. Specifically, in a study of patients with solid tumors receiving GD2-targeting CAR-T cells (with 4-1BB and CD3ζ signaling domains), the frequency of naive T cells in the pre-manufacturing apheresis product correlated with in vivo CAR-T expansion [20]. Similarly, in patients with large B-cell lymphoma receiving axicabtagene ciloleucel (CD19-targeting CAR with CD28 costimulatory domain), higher naive T cell content in the infused product was associated with greater peak expansion [21]. Additionally, in patients with relapsed/refractory B-cell lymphoma receiving tisagenlecleucel (CD19-targeting CAR with 4-1BB costimulatory domain), a higher proportion of naive T cells in the apheresis product was associated with improved clinical responses [22]. Naive T cells exhibit greater proliferative potential and the ability to differentiate into memory subsets during in vitro culture, thereby generating more durable effector cells [23, 24]. Similarly, a favorable CD4/CD8 ratio is linked to better clinical outcomes in patients with B-cell neoplasias receiving CD19-targeting CAR-T therapy [25]. We acknowledge that the baseline CD4/CD8 ratio in peripheral blood may not directly reflect the composition of the final CAR-T product, but rather represents the host’s immune fitness. Nevertheless, this pre-apheresis systemic immune status may influence the quality of the harvested T cells and subsequent in vivo expansion capacity. Furthermore, a retrospective study of CD19-targeted CAR-T therapy in B-cell non-Hodgkin lymphoma demonstrated that patients with CD4+ helper T-cell recovery exceeding 200/μL had significantly prolonged median PFS (20.8 months vs. 1.7 months, p < 0.0001) [26]. Our results are consistent with these observations, raising the possibility that baseline immune characteristics could influence not only product attributes but also in vivo expansion kinetics and, potentially, survival outcomes. However, given the borderline statistical significance and limited sample size, these findings should be considered hypothesis-generating rather than conclusive. The role of Tregs in CAR-T therapy is complex. Tregs are potent suppressors of T-cell responses and may limit CAR-T expansion and antitumor activity [27, 28]. In this study, we found that activated Tregs, rather than total Tregs, showed a trend toward association with poor prognosis, a distinction that may have biological relevance. However, this finding was of borderline statistical significance (P = 0.057) and requires validation in larger studies before any clinical implications can be drawn. Activated Tregs represent the most functionally potent subset within the Treg pool and exhibit stronger suppressive function. They may limit CAR-T expansion through direct contact-dependent mechanisms or indirectly via competitive consumption of cytokines such as IL-2, suggesting that the host immunosuppressive state may influence CAR-T efficacy even before treatment initiation [29, 30]. However, given that the association between activated Tregs and PFS was of borderline statistical significance and the sample size was limited, these findings should be considered exploratory and warrant validation in larger prospective cohorts. Future studies should further delineate the functional differences among distinct Treg subsets (eg, resting Tregs, activated Tregs, tumor-infiltrating Tregs) and their mechanisms of influence on CAR-T efficacy. In exploratory subgroup analyses, we examined whether the association between baseline PD-1 expression and clinical outcomes differed according to receipt of PD-1 inhibitor maintenance therapy. Because of the limited sample size (n = 11 in the non-maintenance group and n = 22 in the maintenance group), no statistically significant associations were observed in either subgroup (log-rank p = 0.55 and p = 0.24, respectively). Nevertheless, a directional trend was noted: among patients who received PD-1 inhibitor maintenance therapy, those with high baseline PD-1 expression showed a numerically longer PFS (with the survival curve consistently above that of the low PD-1 group), whereas no such trend was observed in the absence of maintenance therapy. This observation is hypothesis-generating only and should not be interpreted as evidence of a true interaction. Notably, patients receiving PD-1 inhibitor maintenance therapy were enriched for high-risk features (≥3 prior lines or bulky disease), a subgroup typically expected to have the poorest outcomes. This selection bias limits direct comparisons between the two groups. One speculative interpretation, which requires prospective validation, is that high baseline PD-1 expression might reflect a state of pre-existing T-cell exhaustion that could be partially reversed by PD-1 checkpoint blockade administered after CAR-T infusion [13]. Previous studies have shown that PD-1 inhibitors can reverse the immunosuppressive effects mediated by the IL4I1-induced IDO-AHR-kynurenine metabolic pathway, thereby restoring the cytotoxic function of CAR-T cells [31]. Similarly, in patients receiving sintilimab as consolidation therapy after CD19-targeted CAR-T treatment, median progression-free survival was significantly prolonged (42.9 months vs. 3.0 months) [15]. However, it is important to emphasize that assignment to PD-1 inhibitor maintenance therapy in the present study was non-randomized and subject to selection bias; patients with high-risk features were more likely to receive combination therapy based on institutional practice. Although measured baseline characteristics were balanced between the two groups (Supplemental Table 1), the possibility of residual unmeasured confounding and selection bias cannot be excluded. Given the non-randomized assignment, the lack of statistical significance, and the small subgroup sample sizes, we cannot exclude the possibility that the observed directional trends are due to chance, selection bias, or residual confounding. Therefore, these findings do not support any clinical recommendation regarding the use of baseline PD-1 expression as a predictive biomarker. Well-powered, prospective randomized controlled trials are needed to determine whether baseline PD-1 expression can identify patients who may benefit from post-CAR-T PD-1 inhibitor maintenance therapy. This study has several limitations. First and most importantly, the small sample size (n = 33) substantially limits statistical power, rendering definitive conclusions impossible. Most associations either showed only borderline statistical significance (P = 0.05–0.10) or did not remain significant after FDR adjustment. Second, although we provided a comparison of baseline clinical variables between the low-expansion and high-expansion groups (Supplemental Table 1) and conducted an exploratory limited multivariable Cox regression analysis adjusting for the IPI score (Supplemental Table 2), these adjustments did not incorporate other key potential confounders, including ECOG PS, number of prior lines of therapy, bulky disease, and lymphodepletion intensity. Therefore, the reported hazard ratios may still be subject to residual confounding and should not be interpreted as adjusted effect estimates. Third, assignment to PD-1 inhibitor maintenance therapy was non-randomized and patients with high-risk features (≥3 prior lines or bulky disease) were preferentially assigned to receive combination therapy, introducing selection bias. Although measured baseline characteristics appeared balanced between the two groups (Supplemental Table 1), the possibility of residual unmeasured confounding and selection bias cannot be excluded, and the subgroup sample sizes (n = 11 vs. n = 22) were too small for meaningful comparative analyses. Fourth, as a retrospective, single-center study, the generalizability of the findings is limited. Fifth, the follow-up duration was relatively short, precluding assessment of long-term remission durability. Given these inherent constraints, all findings reported herein should be considered exploratory and hypothesis-generating only. They are not intended to provide definitive evidence of association or causation and must be validated in larger, prospectively designed cohorts. Additionally, although GBTM was applied to avoid arbitrary cutoffs, the limited sample size (n = 33) prevented stable convergence of 3-group and 4-group models. This technical limitation inherently supports the selection of the 2-group solution, which demonstrated excellent classification accuracy. The binary classification (low-expansion vs. high-expansion) is also clinically interpretable and consistent with prior literature [20]. Despite these limitations, this study has several methodological strengths that support its exploratory findings. The application of GBTM avoided the subjectivity associated with classification based on arbitrary cutoffs. Comprehensive immune profiling provided a detailed view of baseline immune status. All patients received the same CAR2219 therapy, and expansion monitoring was standardized. Nonetheless, these strengths do not overcome the fundamental limitations of small sample size and exploratory design. In conclusion, this exploratory, hypothesis-generating study identified preliminary associations between baseline immune cell composition—particularly naive T cells, helper T cells, the CD4/CD8 ratio, and activated Tregs—and CAR-T expansion as well as clinical outcomes in patients with R/R LBCL. These findings should not be interpreted as supporting immediate clinical application but rather support further prospective studies evaluating the utility of baseline immune profiling for risk stratification in CAR-T therapy. Furthermore, exploratory subgroup analyses raised a speculative hypothesis that the direction of association between baseline PD-1 expression and PFS may differ according to receipt of PD-1 inhibitor maintenance therapy—a hypothesis that warrants systematic evaluation in prospective randomized controlled trials before any clinical consideration. Author contributors Data acquisition: Xin-di Liu, He-nan Wang, Li-jie Zeng, Yuan-zheng Liang, Fu-li Li, Jin-yan Zhu, Jia Cong, Liang Wang. Quality control of data and algorithms: Yuan-zheng Liang, Fu-li Li. Data analysis and interpretation: Xin-di Liu, He-nan Wang, Jia Cong. Statistical analysis: Xin-di Liu, Jia Cong. Manuscript preparation: Xin-di Liu, He-nan Wang, Liang Wang. Manuscript editing: Xin-di Liu, He-nan Wang, Li-jie Zeng, Yuan-zheng Liang. Manuscript review: Jia Cong, Liang Wang. All authors read and approved the final article. Liang Wang is the guarantor of the study. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary material 1 Supplementary material 2 Supplementary material 1 Supplementary material 2