Implications for practice

A nomogram was developed to predict the therapeutic response of the CD19 CAR-T infusion product to lymphoma. Based on the receiver operating characteristic curves and the calibration curves, this nomogram identifies several variables with great predictive power for CD19 CAR-T infusion product responses to lymphoma, which provide the basis for clinical adjustment of the treatment plan.

Introduction

Chimeric antigen receptor T cell (CAR-T) therapy is different from conventional small molecule or biological therapy. It collects T cells from patients’ blood through single cell collection. It can express the chimeric antigen receptor (CAR) to recognize and attack tumor cells expressing specific antigens and after modification. At present, its efficacy in diffuse large B cell lymphoma (DLBCL) has been confirmed [1]. The US Food and Drug Administration (FDA) had approved three CD19 CAR-T products, axicabtagene ciloleucel, tisagenlecleucel and lisocabtagene maraleucel to treat refractory/relapsed B cell non-Hodgkin’s lymphoma (r/r B-NHL). Research has shown that the total effective rate (ORR) of treatment is 52 to 82%, and the complete effective rate (CR) is 40 to 58% [2, 3]. However, some patients are not sensitive to CAR-T therapy, and for those who are effective with CAR-T therapy, studies have found that 50% of patients experience recurrence eventually [4]. This is partly due to the heterogeneous influence of functional T cells, tumor microenvironment [5] and patient baseline status. In addition, early clinical response evaluation is also important to improve the cost-effectiveness due to the high cost of CAR-T therapy. Therefore, an early prediction model of treatment response is crucial for patients.

The factors that influence the clinical response to CAR-T therapy are not fully understood. Several studies have attempted to determine the impact of these clinical factors [5,

Results

Clinical characteristics of the study population

Table 1 shows the patients’ baseline characteristics and the CD19 CAR-T cells they initially received. A total of 43 patients with r/r NHL were enrolled in training cohort, comprising 36 (83.7%) cases of diffuse large B cell lymphoma, 2 (4.7%) cases of mantle cell lymphoma, 2 (4.7%) cases of follicular lymphoma, 2 (4.7%) cases of B cell lymphoblastic lymphoma and 1 (2.2%) case of Burkitt’s lymphoma. There were approximately three times as many men as women, and the median age was 56 years (range 26–85 years). Moreover, ten patients (23.3%) had bone marrow involvement prior to CAR-T cell infusion, and 23 (62.7%) patients had a high tumor burden. Meanwhile, the median number of lines of therapy prior to CAR-T cell infusion was 3 (range, 1–7).

The median dose of infused CAR-T cells was 3.82 × 10^6/kg (range, 0.85–12.79). Meanwhile, the median WBC, hemoglobin, platelet, LDH, CRP and ferritin counts before lymphodepletion (baseline blood count) were 3.97 × 10^9/L, 109 g/L, 154 × 10^9/L, 225.3U/L, 4.52 mg/L and 379 ng/ml. Before treatment, the median CD4/CD8 ratio of peripheral blood T cells was 1.1 (0.15–13.43), the median proportion of Treg cells was 5.71% (0.07–31.47), the proportion of Tcm in Th and Tc cells was 7.68% (0.38–62.95) and 31.88% (3.17–64.77), respectively. The proportion of Tn in Th and Tc cells was 13.55% (1.15–46.77) and 8.64% (0–48.68), respectively. The proportion of Teff in Th and Tc was 2.88% (0.16–50.38) and 33.43% (0–79.28), respectively. The proportion of Tem in Th and Tc cells was 34.11% (5.72–92) and 35.12% (0–84.97), respectively. Thirty-one (72.1%) patients achieved CR or PR one to three months after CAR-T cell infusion. Additionally, the patients characteristics of validation cohort are also presented in Table 1, including 14 DLBCL and 1 FL. To further explore the prognostic value, the long-term survival was compared between the ratio of Tcm in Tc and Tn in Tc. The median follow-up of this study was 9 months (1.37–74.83). In the training cohort, the difference between the proportion of Tcm in Tc (log rank p = 0.88) and Tn in Tc (log rank p = 0.96) does not cause the comparison difference of PFS (Supplementary Fig. 1).

Factors associated with CR and PR in r/r NHL patients after CAR-T therapy

To search for possible factors for CD19 CAR-T treatment in r/r NHL patients, we first performed univariate logistic regression analysis on patients’ clinical characteristics (Table 2), baseline hematological parameters (Table 3) and the first infusion of CAR-T cells (Supplementary Table 2). The results of univariate analysis showed that the patient’s baseline tumor burden, ECOG score, the proportion of Treg cells in peripheral blood T cell subsets and the proportion of Tcm and Tn in Tc cells at diagnosis were significantly associated with remission (CR and PR) after CAR-T cell infusion (p < 0.05). As demonstrated in Fig. 3B–D, the proportion of Treg cells in the response group was significantly lower than in the non-response group (SD and PD). On the contrary, the proportion of Tcm and Tn in Tc cells was significantly higher than that in the non-response group. Although age and the proportion of Tn in Th cells shown in Fig. 3E present correlation with remission, the difference was not statistically significant. However, in Fig. 3A, the CD4/CD8 ratio was not different between the two groups, which may be related to the small sample size.

Table 2 Univariate logistic regression analyses of the clinical characteristics of r/r NHL patients associated with CR or PR
Full size table
Table 3 Univariate logistic regression analyses of the blood index of r/r NHL patients associated with CR or PR
Full size table
Fig. 3
figure 3

The proportions of T cell subsets in NHL patients. A There was no significant difference in the proportion of CD4 + and CD8 + T cells between response and non-response group (p = 0.6268). B The proportion of Treg cells in the response group was significantly lower than in the non-response group (p < 0.05). C, D The proportion of Tcm and Tn in Tc cells was significantly higher than that in the non-response group (p < 0.05). E The proportion of Tn in the response group was higher than that in the non-response group, but the difference was not statistically significant (p > 0.05)

Full size image

Identification of independent factors influencing response to CAR-T therapy

To find out more about the independent factors for response to CD19 CAR-T therapy, the statistically significant factors from the above univariate analysis (univariate logistic p 0.05) were added to the binary logistic regression analysis (Supplementary Table 3). The results of the binary analysis showed that the changes in the proportions of T cell subsets collected from the patients’ peripheral blood before CD19 CAR-T infusion were significant independent factors for the therapeutic effect of CAR-T cells, among which the proportions of Tcm and Tn in Tc cells significantly influenced remission (p < 0.05) (Fig. 4A). ROC curve results showed that the AUC of Tcm in Tc and Tn in Tc for predicting the efficacy of CAR-T cell therapy was > 60%, the AUC of Tcm in Tc was 0.855 (95%CI 0.737–0.9730), and the AUC of Tn in Tc was 0.836 (95%CI 0.712–0.960) (Supplementary Table 4) (Fig. 4B).

Fig. 4
figure 4

Discriminative ability of Tcm and Tn in Tc between CR and NR. A Multivariate logistic regression analysis showed that Tcm and Tn in Tc were closely related to CAR-T treatment response (p < 0.05). B The ROC curve showed the ability of the T subsets to discriminate between the response and non-response groups. Tcm and Tn in Tc represented good discriminative power with AUC > 0.6, respectively

Full size image

Establishment and validation of an early prediction model for CD19 CAR-T treatment response

Potential biomarkers with concurrent AUC > 0.6 were identified based on the selection criteria of p < 0.05 between the remission and non-remission groups. We therefore selected Tcm in Tc cells and Tn in Tc cells as predictive biomarkers of early response after CD19 CAR-T therapy. Based on the baseline percentage of Tcm and Tn in Tc cells, we developed an early prediction model of the treatment effect to analyze the likelihood of remission. Then, to assess remission response in clinical practice, we developed a nomogram that could help predict the likelihood of CR and PR for each individual in clinical practice. Finally, we internally validated the original data and drew a calibration curve to further evaluate the authenticity and accuracy of the prediction model in the actual situation. The results showed good agreement between the likelihood of predicting remission and the observed early response (Figs. 5A and B). The total scores of Tcm and Tn were calculated in the nomogram, and the ROC curve was drawn to predict the treatment effect. The AUC result was 0.914 (95%CI 0.832–0.996), showing good predictive ability (Fig. 5C). Moreover, we further validated this predictive model in an independent cohort. The internal validation cohort consisted of 15 r/r B-NHL patients enrolled during a subsequent period of time. The baseline characteristics of the validation group are shown in Supplementary Table 1. According to the established predictive models for response, ROC curve results of validation cohort showed that the AUC of Tcm in Tc and Tn in Tc for predicting the efficacy of CAR-T cell therapy was > 60%, the AUC of Tcm in Tc was 0.860 (95%CI 0.6587–1.0000), and the AUC of Tn in Tc was 0.940 (95%CI 0.8137–1.0000) (Supplementary Fig. 2A). The ROC curve was drawn to predict the validation cohort treatment effect, and AUC result was 0.9301 (95%CI 0.8568–1.0), showing good predictive ability (Supplementary Fig. 2B). The results showed that the prediction model had high sensitivity and specificity in predicting the effect of CD19 CAR-T treatment (sensitivity 83%, specificity 74.2%), and the maximum Youden index was 0.742, and the prediction results were in good agreement with the actual situation (kappa, 0.616), indicating that the model has good predictive ability and has some clinical application value.

Fig. 5
figure 5

The early prediction model nomogram and calibration curve. A A nomogram was created based on the proportion of Tcm and Tn in Tc to predict response to CAR-T treatment. B Calibration curves of the early predictive model comparing the predicted probability of response. C The total scores of Tcm and Tn were calculated by the nomogram prediction model, and the ROC curve was drawn to predict the response to CAR-T therapy. AUC = 0.914 (95%CI 0.832–0.996)

Full size image

Discussion

Chimeric antigen receptor-modified T (CAR-T) cells have shown significant advantages in the treatment of hematological malignancies. The clinical trial data of our center have good therapeutic effects and are similar to the clinical effects of commercial CAR-T products [14]. However, some patients are not sensitive to CAR-T therapy, and for those who are effective with CAR-T therapy, studies have found that 50% of patients experience recurrence eventually. In this retrospective study, we analyzed patients’ clinical characteristics, baseline hematological parameters and the proportion of T cell subsets isolated from peripheral blood before treatment to identify independent risk factors affecting the efficacy of CAR-T cell therapy. Combined with univariate and multivariate analyses, it was found that the proportions of Tn and Tcm in patients’ peripheral blood CD8 + T cells were positively correlated with clinical outcome after CAR-T treatment. Therefore, we established an early prediction model to analyze the possibility of CR and PR in lymphoma patients based on the proportions of Tcm in Tc and Tn in Tc. This model helps to predict early treatment response in patients with r/r B-NHL. Based on this model, identification of T cell phenotypes prior to CAR-T manufacturing may lead to early intervention by enriching dominant cell subsets or consuming undesirable cell populations or functional states during manufacturing, thereby reducing the failure and relapse rates of CAR-T therapy in r/r NHL.

With the advancement of CAR-T treatment technology, the target population for its treatment is constantly expanding. CAR-T treatment has moved from being considered a third-line treatment or greater to a second-line treatment. With the expansion of the target population for CAR-T treatment, there will inevitably be differences in the responses of different populations to CAR-T treatment. If biomarkers can be used to distinguish them, it can help select more suitable target patients or combine other methods to improve the treatment effect for patients with potential poor efficacy.

In this study, the results of univariate analysis showed that the patient’s baseline tumor burden and ECOG score were significantly associated with remission (CR and PR) after CAR-T cell infusion (p < 0.05). Our results are basically consistent with other research findings. Vercellino et al. [8] analyzed the characteristics of 116 patients at the time of decision (TD) to use commercial CAR (axicabtagene ciloleucel, n = 49; tisagenlecleucel n = 67) and at the time of treatment (TT) and found that the risk factors for early progression of TD and TT during treatment were extranodal (EN) site involvement (≥ 2 sites) and lymphoma burden (LDH, TMTV). This reminds us that disease control before administering CAR-T cells is crucial for better response. Some biomarkers mainly choose to exclude patients who have many potential negative factors that may affect the efficacy of CAR-T through a reverse thinking approach. The research team retrospectively analyzed 34 paired samples in a dual-target CAR-T clinical study and found that 38.2% of patients had low expression of NOXA before CAR-T treatment. Patients with low expression of NOXA received CAR-T treatment, and their PFS and OS were significantly reduced compared to those with high expression of NOXA. ORR and CR rates were also affected [15]. Interestingly, the team’s research found that low expression of NOXA can be reversed by HDACi inhibitors. The HDACi inhibitor Chidamide, independently developed in China, can enhance the killing activity of CAR-T against tumor cells by reversing the low expression of NOXA to high expression [15]. The reversible biomarker provides a foundation for subsequent clinical conversion.

The cellular and molecular diversity of CAR-T cell products is an important factor in the efficacy of CD19 CAR-T cell therapy in LBCL. In this study, we found that T cell heterogeneity in blood samples isolated for CAR-T product manufacturing had a profound impact on clinical outcomes. The high proportion of Tn and Tcm in CD8 + T cells was associated with an improved treatment response. This is consistent with Lamure et al. [16] who suggested that Tisa-cel response was associated with expansion of central memory CD8 + T populations, and Locke et al. [17] described results on the ZUMA-1 cohort that high percentages of naïve-like (CCR7 + CD45RA +) T cell were associated with a higher response rate. Deng et al. [18] found that product cells transfused into patients with durable CR at 3 months were rich in CD8 memory T cell phenotypes compared to PR/PD patients. And Zhang et al. [19] found that knocking out BATF shifts the population toward a more central memory subset and enhances the antitumor activity of CAR-T cells against solid tumors. CD19 CAR-T cells from completely responding individuals with chronic lymphocytic leukemia were enriched in memory-related genes, whereas CAR-T cells from non-responders upregulated effector- and exhaustion-related genes, suggesting that the quantity of Tcm cells is one of the major parameters for the persistence and activity of CAR-T cells [20, 21]. In addition, Lamure et al. [16] found that an increase in Treg cells appeared to be associated with non-response to treatment and could drive clinical relapse. In our study, the proportion of Treg cells was indeed negatively correlated with clinical outcome, although no statistically significant difference was found in the multivariate analysis, which was speculated to be due to the small sample size included, and further clarification is needed in larger cohort studies. This is similar to the finding for the CD4/CD8 ratio. Some investigators believe that the CD8 + T cell-mediated immune response to transgenic CAR-T cells after CAR-T cell infusion limits the persistence of CAR-T cells and increases the risk of relapse in some patients [22]. This study found that the proportion of CD4 + T cells was slightly higher than that of CD8 + T cells in the remission groups. Although this difference was not statistically significant, it also shows that the CD4/CD8 ratio may have an impact on clinical outcome. However, this study has some limitations. First, the number of patients is relatively small, and the predictive model we constructed needs to be further validated in larger prospective studies. In addition, most patients were discharged within 1–3 months after CAR-T treatment, and we only observed short-term efficacy after CAR-T treatment, not long-term outcomes. Articles indicate a positive correlation between the Tn or Tcm ratio and the persistence of CAR-T cells in vivo. There is a high proportion of Tn and memory T cell populations in the patient cohort with CAR-T cells persisting for more than 6 months and sustained remission [23]. We analyzed that our training cohort lasted for 3 years, so there was no difference in the observed results of PFS. Meanwhile, our study is limited to blood, and future studies and analyses of lymph node biopsy should be included to expand the content of this prediction model to comprehensively and deeply analyze the clinical outcomes of patients.