1 Introduction
Chimeric antigen receptor (CAR)-T cell therapy has become an important anti-lymphoma approach, in which autologous T cells are genetically engineered
in vitro, then infused back into the patients to target tumor-associated antigens. The clinical efficacy of anti-CD19 CAR-T cell therapy has been proven in both clinical trials and real-world practice for relapsed or refractory diffuse large B cell lymphoma (DLBCL) [
1-
3], with 40%–50% of patients achieving long-term remission [
1,
2]. Thus far, two major patterns of resistance to CAR-T cell therapy have been described. The first pattern, defined as primary resistance, is due to the lack of response to CAR-T cells, resulting from impaired death receptor signaling and dysfunctional CAR-T cells [
4]. The second pattern, defined as secondary resistance, refers to disease progression after response, mainly due to loss of CD19 antigen [
5,
6] and low quality of CAR-T cell expansion [
7,
8]. However, currently, the interactions of CAR-T cells with the tumor microenvironment and its effect on CAR-T cell therapy remain largely unknown.
Our previous study indicated that tumor-associated macrophage infiltration could reduce the response to anti-CD19 CAR-T cell therapy [
9]. To further investigate the influence of tumor microenvironments on lymphoma progression, we conducted a phase I/II trial of CD28z CAR-T cell therapy (FKC876, ChiCTR1800019661), and performed single-cell RNA sequencing (scRNA-seq) on tumor samples collected during therapy from DLBCL patients.
2 Methods
2.1 Study design and participants
Data were sourced from the phase I/II trial of CD28z CAR-T cell therapy (FKC876, ChiCTR1800019661) and the previous phase I trial of 41BBz CAR-T cell therapy (JWCAR029, NCT03355859) [
9]. Full eligibility and exclusion criteria of FKC876 are described in the Supplemental Methods. The study was approved by the Review Board of Ruijin Hospital and informed consent was obtained in accordance with the
Declaration of Helsinki.
2.2 Treatment and evaluation
The safety and efficacy of CAR-T cell therapy for relapsed or refractory CD19-positive non-Hodgkin lymphoma patients were assessed. Regimens of fludarabine and cyclophosphamide lymphodepletion preconditioning were initiated on day 5 and completed by day 3 before CAR-T cell infusion. CD28z CAR-T cells of dosage 2 × 106 cells per kg body weight were administered intravenously. Blood cell counts, serum biochemistry, coagulation panels, and cytokine levels were monitored for 29 days after infusion, and followed up for 24 months thereafter. Primary tumor response was assessed on day 29 and also followed up for 24 months. The primary endpoint was the incidence of adverse events (AEs). Secondary endpoints included pharmacokinetics, complete remission (CR) rate, duration of response, progression-free survival (PFS), progression-free survival ratio (previous line of treatment/CAR-T), and overall survival (OS). Details on the preparation of CAR-T cell products are described in the Supplementary Methods.
2.3 CAR-T cell detection in vivo
Blood samples were obtained from patients before and at pre-determined intervals after CAR-T cell infusion. CAR-T cell levels were assessed by quantitative real-time polymerase chain reaction (RT-PCR) in the CD28z CAR-T study and by flow cytometry in the 41BBz CAR-T study [
9]. The CAR-T pharmacokinetics study began with CD28z and 41BBz CAR-T cell infusion (
in vivo amplification and persistence) and continued until CAR-T cells were no longer detected by quantitative RT-PCR for CD28z CAR-T cells and by flow cytometry for 41BBz CAR-T cells.
2.4 Cytotoxic potency, macrophage differentiation, and T cell suppression assay
The cytotoxic potency of CAR-T cells was assessed by calcein release assays. M2 macrophages were differentiated from monocyte using cytokines CSF, IL-4, IL-6, and IL-13. T cell suppression assay was performed using CFSE-labeled T cells to coculture with/without M2 macrophages and then the suppressive capacity was calculated. Details on the preparation of these experiments are described in the Supplementary Methods.
2.5 T cell transduction and culture conditions
Healthy donor T cells were selected from healthy donor PBMCs by a pan-T cell selection kit (Miltenyi) following the manufacturer’s instructions. The cells were then activated, expanded for 24 h using CD3/CD28 beads (Dynabeads, Gibco), transduced 24 h later with the lentivirus, and incubated in RPMI 1640 medium with 10% fetal bovine serum (FBS) and IL-2 (300 IU/mL) for two days and then CAR-T cells were expanded in the RPMI 1640 medium with 10% FBS and IL-2 (300 IU/mL).
2.6 CAR19 cloning and lentivirus production for in vitro study
DNA fragments encoding the FMC63 (anti-CD19) [
10] were synthesized (Genescript) and subcloned into pLJM1(Addgene) in frame with the sequence encoding the IgG1 hinge, the transmembrane domain of 4-1BB, or CD28 with intracellular domains of 4-1BB and CD3z under the EF1a promoter, as shown in Fig. S1. To produce lentiviral particles for transduction into T cells, HEK293T cells were transfected with the vector carrying CAR19 together with the packaging plasmids psPAX2 and pMD2.G (Addgene). Thereafter, viral supernatants were collected after 48 h of incubation.
2.7 Single-cell RNA sequencing (scRNA-seq)
Tissue biopsies were collected on day −6 (one day before lymphodepletion), on day 9 after CAR-T cell infusion when adequate CAR-T cell expansion was already detected, and at disease progression, if existed. Collected samples were preserved at 2–8 °C, then washed and minced prior to tissue dissociation, digestion, and purification. Removal of red blood cells and cell count were then carried out to complete the single-cell suspension. Concentration calibration was set to 1 × 105 cells/mL in Phosphate Buffered Saline for the single-cell suspension before being loaded onto a microfluidic chip for scRNA-seq libraries to be constructed and sequenced. Raw reads were then processed to generate gene expression matrices. Reads with the same cell barcode, UMI, and gene were grouped together to generate the number of UMIs per gene per cell. The number of cells was then determined based on the inflection point of the number of UMIs versus the sorted cell barcode curve. Details on the scRNA-seq preparation and analysis were described in the Supplementary Methods.
2.8 Immunohistochemistry
Immunohistochemistry was performed on 5-µm paraffin sections by the indirect method (EnVision) using primary antibodies against CD163, IL1B (BJZS-Bio, ZM-0428, Abcam, ab2105), and anti-rabbit- or anti-mouse-IgG as secondary antibodies (Biocompare, GV809, GV821). For absolute quantification, five randomly selected HPFs (×200) were analyzed in each section.
2.9 Statistical analysis
The proportion of patients with an objective response was summarized with descriptive statistics. Time-to-event analyses for the duration of response, PFS, and OS were assessed with the Kaplan–Meier method, and 95% CIs for responses were calculated with the Clopper–Pearson method. PFS was calculated from the date when treatment was initiated to the date when disease progression or death was recognized or the date of the last follow-up. OS was calculated from the date when treatment began to the date of death or last follow-up. All data were tested for normal distribution (D’Agostino–Pearson omnibus normality test) before analyses. If the data were normally distributed, a t-test or one-way ANOVA was used. For non-normally distributed data, a nonparametric test (Wilcoxon signed-rank test or Friedman test) was used (GraphPad Prism v7). Immune cell populations were identified by the CIBERSORT algorithm (v 1.03) with its reference list. All comparisons used a two-sided α of 0.05 for significance testing.
3 Results
3.1 Patients’ characteristics
As shown in Fig.1, 34 patients were screened, while 27 patients were eventually enrolled. Three patients were not eligible for reinfusion: one patient experienced cell preparation failure, and 2 patients experienced disease progression. Overall, 24 patients were included in the trial. Two patients were excluded from the efficacy analysis: one patient abandoned treatment on day 17, and another patient died from gastrointestinal bleeding on day 20. Patients’ characteristics are summarized in Table S1. Patients had a median age of 56 years (ranging from 29 to 63). Among 24 patients, 58.3% of patients had stage III or IV disease, 25% of patients had bulky disease (≥ 7.5 cm), and 58.3% of patients were categorized into intermediate- and high-risk International Prognostic Index groups. Patients received a median of three prior lines of therapy (range 1–5). Two patients relapsed after autologous stem cell transplantation.
The most common AEs of any grade observed were pyrexia (24/24, 100%), leukopenia (24/24, 100%), and neutropenia (24/24, 100%). The most common AEs of Grade 3 or higher were leukopenia (23/24, 95.8%), neutropenia (23/24, 95.8%), and thrombocytopenia (13/24, 54.1%). Grade 1 cytokine release syndrome (CRS) occurred in all patients. Grade 2 neurotoxicity occurred in one patient, while Grade 3 neurotoxicity occurred in one patient, who had intermittent epilepsy. CRS and neurotoxicity were found to be self-limiting and reversible. Eighteen patients received tocilizumab (8 mg/kg), and 12 received glucocorticoids for the management of CRS, neurologic events, or both (Table S2).
3.2 Clinical efficacy
The date of data cut-off for analysis was August 1, 2021. Among 22 evaluable DLBCL patients, the median follow-up was 23.6 (range 1.6–32.6) months. Seventeen patients achieved an objective response on day 29 after CAR-T cell infusion, including seven patients with CR, 10 patients with partial remission, and four patients with stable disease. The proportion of patients with PFS at 24 months was 62.5% (95% CI 29.0–96.0) among those with complete remission at day 90, and 14.3% (95% CI 0.0–32.7) among those with non-CR at day 90. The proportion of patients with OS at 24 months was 100% (not reached (95% CI not estimable–not estimable)) among those with complete remission at day 90, and 35.7% (95% CI 10.6–60.8) among those with non-CR at day 90. In terms of complete remission status at day 90, the median PFS of the CR patients was significantly longer than those of non-CR patients (not reached (95% CI not estimable–not estimable) vs. 3 months (1.4–4.5), HR 0.20, 95% CI 0.07–0.56, P = 0.0033, Fig.1). The median OS of the CR patients was significantly longer than those of non-CR patients (not reached (95% CI not estimable–not estimable) vs. 7.5 months (0.0–37.0), HR 0.15, 95% CI 0.04–0.58, P = 0.0058, Fig.1). Thus, CR status at day 90 was used as the reference time point for further studies.
3.3 Pharmacokinetics
Among the 22 patients, comparing CR with non-CR patients in the CD28z CAR-T trial, CR status at day 90 and the area under the curve in the first 29 days after CAR-T cell infusion showed no significant difference (Fig. S2A and S2B). Furthermore, between CD28z and 41BB, peak value and peak day of CAR-T cells also showed no significant difference (Fig. S2C and S2D). The daily growth rate before the peak value of CAR-T cells was significantly higher in CR patients (3.0-fold change) than in progressed patients (1.7-fold change) of the 41BBz CAR-T trial. Such a difference was not observed in the CD28z CAR-T trial (Fig.2). The persistence of CAR-T cells were also longer in the 41BBz CAR-T trial than in the CD28z CAR-T trial (Fig. S2F).
3.4 Gene mutation pattern
A gene mutation panel of 55 genes related to oncogenesis of DLBCL [
11] was performed on tumor samples of 21 patients (six CR and six progressed patients at day 90 of the CD28z CAR-T trial, as well as five CR and four progressed patients at day 90 in the 41BBz CAR-T trial). No significant difference in gene mutation pattern was observed between the CR and the progressed patients (Fig. S3).
3.5 Tumor microenvironments before CAR-T cell therapy
To better understand the impact of the tumor microenvironments on CAR-T cell therapy, we performed scRNA-seq on tumor samples collected from patients during treatment and follow-up. We collected scRNA-seq data from one patient with long-term CR, and another patient with early progression. According to the signature genes of the immune cells (Fig. S4), uniform manifold approximation and projection (UMAP) showed that cell clusters overlapped between the CR patient and the progressed patient before CAR-T cell therapy (Fig.2). Distinct clusters were subsequently analyzed (Fig.2), including B cells, macrophages, T cells, and fibroblasts, according to the signature genes MS4A1, CD68, CD3D, and DCN, respectively. B cells were the major cell type before CAR-T cell therapy in both the CR (91.4%) and the progressed patient (82.7%) (Fig.2). Among the other cell types, the percentages of macrophages and T cells were higher in the progressed patient (11.2% and 5.3%) than those of the CR patient (6.6% and 1.4%).
Previously, we reported that macrophages influence the outcome of CAR-T cell therapy [
9]. Using UMAP to further analyze specific subtypes of macrophages, we found 10 subclusters with distinct profiles from the CR and the progressed patients before CAR-T cell therapy (Fig.2). Clusters 1, 2, and 4 were defined as M2-subtype macrophages according to the signature genes
CD163 and
MRC1, while clusters 0, 3, 5, 6, 7, 8, and 9 were defined as M1-subtype macrophages according to the signature gene
IL1B. Notably, the percentage of M2-subtype macrophages was higher in the progressed patient (48.2%) than in the CR patient (29.2%) (Fig.2). These data suggested that pre-existing immunosuppressive tumor microenvironments, particularly the presence of M2-subtype macrophages, were associated with the remission status of CAR-T cell therapy.
3.6 In vitro features of M2-subtype macrophages on CAR-T cells
To provide further evidence that M2-subtype macrophages suppress anti-tumor activity of CAR-T cells, we designed an experiment in which anti-CD19 CAR-T cells and Raji cells were cocultured with or without M2-subtype macrophages. As expected, the cytokine secretion was significantly decreased in both CAR-T CD28z cells and CAR-T CD28/41BBz cells cocultured with M2-subtype macrophages against Raji cells, when compared to the groups cocultured without M2 macrophages (Fig.3 and 3B). Bioluminescence imaging (BLI)-based killing assay was then conducted. Briefly, Raji-firefly luciferase (ffluc) cells were cocultured with CAR-T cells with or without M2-subtype macrophages for six hours in a 96-well plate. The plate was then read using the Lumina imaging system. The results showed that lysed Raji cells significantly decreased when Raji cells were cocultured with CAR-T cells in the presence of M2-subtype macrophages, when compared to those without M2-subtype macrophages at the CAR-T/Raji ratio of 20:1, 10:1, 5:1, and 1:1 (Fig.3), indicating that M2-subtype macrophages inhibited the cytotoxic activity of CAR-T cells to Raji cells.
To examine the effect of M2-subtype macrophages on CAR-T cell proliferation, CAR-T cells were cocultured with M2-subtype macrophages at a 1:1 ratio for 72 h without any cytokine, and the absolute number of CAR-T cells was counted at the end of coculture. Remarkably, with the presence of M2-subtype macrophages in the coculture system, the cell count of CAR-T CD28z cells was significantly decreased when compared to CAR-T (CD28z) cells alone. Similar results were observed in CAR-T CD28/41BBz cells with M2-subtype macrophages (Fig.3). Taken together, M2-subtype macrophages influenced the anti-tumor efficacy of CAR-T cells by decreasing the cytokine production, cytotoxic ability, and proliferation of CAR-T cells.
3.7 Immune cell compartments during CAR-T cell therapy
After assessing the in vitro functions of M2-subtype macrophages, we attempted to understand the role of M2-subtype macrophages in the tumor microenvironments during the CAR-T therapy in vivo. M1/M2-subtype macrophages were classified based on the expression of macrophage signature genes (CD163 and MRC1) for M2-subtype macrophages highly expressed in subclusters 0, 2, and 4, and IL1B for M1-subtype macrophages highly expressed in subclusters 1 and 3 (Fig. S5A). As confirmed by function enrichment analysis of differentially expressed genes, M2 subclusters 0, 2, and 4 were enriched in pathways mediating M2 activation, such as cellular lipid and cholesterol catabolic processes, glycolipid metabolic process, interleukin-10 production, negative regulation of the immune system processes, and positive regulations of epithelial cell migration. In contrast, M1 subclusters 1 and 3 were enriched in pathways mediating M1 activation, such as response to lipopolysaccharide, response to tumor necrosis factor, cellular response to interleukin-1, positive regulation of αβ T cell activation, response to ATP, and nitric oxide biosynthetic process (Fig. S5B). To better illustrate the differences among the different statuses of macrophages, we included the expression of marker genes that contributed to the pseudo-time trajectories (Fig. S5C and S5D). As shown in Fig. S5D, CD80, TLR2, IL1B, IL1A, and ETS2 were related to M1 (clusters 1, 3) polarization, C1QB, C1QA, APOE, CD68, CD164, and PPARG were associated with M2 (clusters 0, 2, 4) activation (Fig. S5C and S5D).
To evaluate the trajectory of macrophages in CAR-T cell therapy, we projected Monocle analysis using scRNA-seq data and located the cells in the trajectory branches according to cell status (single-cell mRNA quantification and differential analysis with census). Notably, the macrophages from the CR and the progressed patients were distributed on the different trajectory branches without any overlapping before CAR-T cell therapy, indicating the macrophages were phenotypically distinct. Macrophages from the progressed patient completely changed their distribution on trajectory branches following the treatment, suggesting a skewed transcriptome profile of macrophages after CAR-T cell therapy, whereas macrophages from the CR patients did not (Fig.4). Subsequently, we tested the compartments of M1- and M2-subtype macrophages at each time point. M2-subtype macrophages decreased in the CR patients from 29.2% before CAR-T cell therapy to 17.2% at CAR-T cell expansion, but it increased in the progressed patient from 51.6% before CAR-T cell therapy to 72.1% at CAR-T cell expansion, and to 85.3% at disease progression. In contrast, M1-subtype macrophages increased in the CR patient from 70.8% before CAR-T cell therapy to 82.8% at CAR-T cell expansion but decreased in the progressed patient from 48.2% before CAR-T cell therapy to 27.9% at CAR-T cell expansion, and to 14.7% at disease progression (Fig.4), as confirmed by immunohistochemistry results (Fig. S6A and S6B). Thus, M2-subtype macrophages of the tumor microenvironments were linked to disease progression upon CAR-T cell therapy, which could not be altered by CAR-T cells.
To explore the biological functions of macrophages on disease progression, we performed gene set enrichment analysis on the different time points of M2-subtype macrophages in the CR and the progressed patients (Fig.4). Notably, metabolic pathways were upregulated, while immune function pathways were downregulated in the progressed patient. In contrast, metabolic pathways were downregulated in the CR patient at CAR-T cell expansion. These data suggest an association of immune cell metabolism with the immunosuppressive tumor microenvironments.
3.8 T cell evolution during CAR-T cell therapy
To further track the evolution of T cells during CAR-T cell therapy, we analyzed scRNA-seq data by generating a tSNE map to identify T cell subclusters according to signature genes, such as T cell subclusters 0 and 6 defined as effector T cells and subcluster 1 as Treg cells (Fig. S7A). As revealed by Monocle analysis, although located on the same trajectory branch before CAR-T cell therapy, T cells displayed distinct features between the CR and the progressed patients after CAR-T cell therapy, especially at disease progression (Fig.5). To track T cell compartments after CAR-T therapy, we evaluated the frequency of effector T and Treg cells. The frequency of effector T cells dramatically increased from 12.5% before CAR-T cell therapy to 61.8% at CAR-T cell expansion in the CR patient; however, in the progressed patient, it only slightly increased from 21.6% before CAR-T cell therapy to 31.8% at CAR-T cell expansion, and to 25.1% at disease progression. In contrast, the frequency of Treg cells decreased from 32.5% before CAR-T cell therapy to 25.6% at CAR-T cell expansion in the CR patient, however, in the progressed patient it increased from 12.3% before CAR-T cell therapy to 30.4% at CAR-T cell expansion, and decreased to 20.9% at disease progression (Fig.5).
In effector T cells, genes, such as CCL5, CCL4, ID2, DUSP2, and RGS2, that activate T cell functions, were highly expressed. For Treg cells, genes, such as FOXP3 and IL2RA, that inhibit T cell functions, were highly expressed (Fig. S7B). Function enrichment analysis of differentially expressed genes indicated that effector T cells were enriched in natural killer cell chemotaxis, positive regulation of lymphocyte chemotaxis, positive regulation of leukocyte chemotaxis, and response to tumor necrosis factor. Treg cells were enriched in negative regulation of T cell proliferation, leukocyte proliferation, and T cell activation (Fig. S7C).
In addition, we performed gene set enrichment analysis of effector T and Treg cells in the CR patient and the progressed patient at different time points (Fig.5). Notably, in the CR patients, natural killer cell chemotaxis and positive regulation of T cell proliferation were upregulated in effector T cells (subclusters 0, 6) at CAR-T cell expansion, whereas for the progressed patient, T cell antigen processing was downregulated at CAR-T cell expansion, and tumor necrosis factor was downregulated at disease progression. However, in the CR patient, natural killer cell chemotaxis was upregulated in Treg cells (subcluster 1) at CAR-T cell expansion but was downregulated for the progressed patient at disease progression. These results indicated that, although effector T cells were present, their anti-tumor functions were suppressed. As expected, the expression levels of LAG3, TIGIT, and PDCD1 were significantly higher on effector T cells from the progressed patient at CAR-T cell expansion and at disease progression. There was no significant difference in the CR patients before CAR-T cell therapy and at CAR-T cell expansion (Fig.5).
3.9 Malignant B cell evolution during CAR-T cell therapy
Furthermore, we analyzed scRNA-seq data by generating a tSNE map to identify malignant B cell subclusters according to the signature genes of B cells (Fig. S8A). The trajectory plots displayed distinct B cell subtypes at different time points for both the CR and the progressed patients. Interestingly, the trajectory of tumor cells in CR patients after CAR-T cell therapy displayed different gene expression profiles, which indicated that CAR-T cells provided pressure on tumor cells and changed tumor cell profile. However, the trajectory of the tumor cells in the progressed patients after therapy was similar to before CAR-T cell therapy (Fig.6). Next, we aimed to understand the mechanisms and analyzed tumor cell subcluster compartments at each time point. Compared to before CAR-T cell therapy, B cell subclusters 0 and 1 were the dominant subclusters in the progressed patient after CAR-T cell therapy and subclusters 3 and 7 were the dominant subclusters in the CR patient (Fig. S8B). Moreover, subclusters 0 and 1 cells were mainly in phase G1/G0 (83.3%) with low levels of proliferation, while most of subclusters 3 and 7 cells were in phase S phase (66.0%) and G2/M phase (31.6%, Fig.6) in the gene synthesis and replication stages. The frequency of subclusters 0 and 1 increased from 34.0% before CAR-T cell therapy to 63.3% at CAR-T cell expansion in the CR patient, but decreased in the progressed patient from 45.3% before CAR-T cell therapy to 35.5% at CAR-T cell expansion, and to 43.3% at disease progression. In contrast, the frequency of subclusters 3 and 7 decreased from 34.3% before CAR-T cell therapy to 8.3% at CAR-T cell expansion in the CR patient, but increased in the progressed patient from 11.5% before CAR-T cell therapy to 34.1% at CAR-T cell expansion, and to 19.5% at disease progression (Fig.6). This was confirmed by function enrichment analysis of differentially expressed genes that subclusters 3 and 7 were enriched in cell cycle pathways (Fig. S8C and S8D).
To investigate the biological functions of tumor cells, we performed gene set enrichment analysis on scRNA-seq data. Both subclusters 0, 1, and subclusters 3, 7 cells showed upregulated hypoxia pathways in the CR and the progressed patients during both CAR-T cell expansion and disease progression. However, subclusters 3 and 7 showed glycolysis, regulation of BCR signaling pathway, and ERK1/2 cascade in the progressed patients after CAR-T therapy till disease progression, but not in the CR patient (Fig.6), suggesting that subclusters 3 and 7 cells were upregulated in glycolytic metabolic activity and BCR signaling activation.
3.10 Cell–cell interactions and relative metabolic disorder at disease progression
To better understand mechanisms related to disease progression after CAR-T cell therapy, we pursued NicheNet (NicheNet: modeling intercellular communication by linking ligands to target genes) analysis to observe the cell–cell interactions by segregating B cells, T cells, M2-, and M1-subtype macrophages from our data sets. We identified more than 20 significant gene interactions. The top 20 ligands that interacted with T cells at disease progression were mostly enriched in macrophages (Fig.7). To assess the transcriptional influence of these ligands, we applied NicheNet to predict ligand-target regulatory potential during T cell evolution. Interestingly, T cell targets were predicted to be influenced by a range of ligands, including APOE, MMP9, FN1, CXCL9, and CCL2, expressed on macrophages. More importantly, those ligands were predicted to promote a variety of genes associated with T cell transcriptome features, such as CD55, VCAM1, SERPINH1, TIMP1, NR4A1, SGK1, MYB, PA2G4, HSPD1, and SPP1 (Fig.7). To confirm that ligands expressed on macrophages promoted targets expressed on T cells, we investigated the scRNA-seq data set from the progressed patient, and observed significantly higher expression levels of APOE, MMP9, FN1, CXCL9, and CCL2 on macrophages at disease progression, as compared to those at CAR-T cell expansion in the progressed patient. Subsequently, we aimed to find the mechanisms of tumor progression after CAR-T cell therapy, and found that the expression levels of four genes (APOE, MMP9, FN1, and CCL2) increased on macrophages at disease progression, as compared to those at CAR-T cell expansion (Fig.7). Interestingly, TIMP1 was the only gene highly expressed on T cells at disease progression, when compared to that at CAR-T cell expansion (Fig.7). These findings are in line with previous studies where TIMP1 was highly correlated with tumor metastasis and increased apoptosis, indicating that macrophages may alter T cell subset by cell–cell interactions.
We found that metabolism-related gene sets in the tumor microenvironments showed significant alternations during CAR-T cell therapy. To better understand the metabolism status of B cells, T cells, M1- and M2-subtype macrophages, we analyzed different metabolic pathways at disease progression, as compared to CAR-T cell expansion. For M2-subtype macrophages, there was an increase in glycolytic rates and medium chain fatty acid catabolic process, and a decrease in response to tumor necrosis factor. For effector T cells, there was a decrease in glycolytic process, medium chain fatty acid catabolic process, alpha-amino and acid metabolic process, and tumor necrosis factor (Fig.7). These findings confirmed that M2-subtype macrophages were able to survive in the tumor microenvironments by increasing the glycolytic process and medium-chain fatty acid catabolic process to suppress the anti-tumor activity of T cells.
4 Discussion
The anti-CD19 CAR-T cell therapy induces 40%–50% long-term remission in relapsed or refractory DLBCL [
3,
12] under the CR at day 90 as a predictive criterion [
1]. T cell quality is related to the clinical efficacy of CAR-T cell therapy, mainly as a dysfunctional phenotype of pre-infusion CAR-T cells and the number of CD45RA
+CCR7
+ T cells and CD8
+ T cells in patients with chronic lymphocytic leukemia and non-Hodgkin lymphoma [
8,
13,
14]. Growing evidence suggests that the immunosuppressive tumor microenvironments may be involved in resistance to CAR-T cell therapy. However, the dynamic changes in the phenotypes and functions of immune cells have not been fully elucidated due to limited access to tissue samples during CAR-T cell therapy. Using cord needle biopsy, we successfully obtained tumor samples from DLBCL patients before CAR-T cell therapy, at CAR-T cell expansion, and at disease progression. Our
in vitro and
in vivo data did not only confirm the essential roles of M2-subtype macrophages in CAR-T cell dysfunction, but also discovered immunosuppressive features of microenvironment components that contributed to disease progression upon CAR-T cell therapy in DLBCL.
Tumor-associated M2-subtype macrophages were significantly associated with poor response to CAR-T cell therapy [
9]. Our scRNA-seq showed that at the single-cell level, M2-subtype macrophages were enriched prior to CAR-T cell infusion in patients who failed to achieve long-term remission. More importantly, tumor microenvironments in the progressed patient showed immunosuppressive properties at CAR-T cell expansion and at disease progression, similar to before CAR-T cell therapy. Therefore, alterations in M2-subtype macrophages could not be counteracted by CAR-T cells and eventually lead to disease progression in DLBCL. To explain the underlying mechanisms, we analyzed the evolution of T cells and tumor cells and found that tumor cells displayed a low-proliferation status after CAR-T cell therapy, and were eliminated by functionally active T cell subsets that expanded after CAR-T cell infusion in the CR patient. In contrast, tumor cells displayed a high-proliferative status, and T cell functions were subsequently suppressed in the progressed patient. CAR-T cell therapy induced cell cycle arrest in tumor cells in the CR patients, but not in the progressed patient. For example, the cytokines produced by CAR-T cells (IFN-γ and TNFα) promote tumor cell growth arrest in G
1/G
0 [
15]. In addition, we found that sustained exposure to immunosuppressive tumor microenvironments induced significant transcriptional alterations, in which
TIMP1 was highly expressed on tumor-infiltrating T cells at lymphoma progression [
16]. Further cell–cell interaction analysis revealed that induction of T cell dysfunction by M2-subtype macrophages could induce persistent immunosuppressive impact on CAR-T cells. Therefore, tumor microenvironment status may be an important indicator of the clinical efficacy of CAR-T cell therapy.
Metabolic reprogramming of the tumor microenvironments is critically involved in T cell function and response to immunotherapy [
17]. Hypoxia, glycolytic process, amino acid metabolic process, and medium chain fatty acid catabolic processes were upregulated in M2-subtype macrophages of the progressed patient during both CAR-T cell expansion and disease progression. These findings are consistent with the report that M2-subtype macrophages rely on oxidative phosphorylation [
18]. Moreover, macrophages are polarized to M2 phenotype by IL10 and TGFβ1, facilitating tumor cell growth, invasion, and metastasis [
19,
20]. Thus, the tumor favorable microenvironment leads to a struggle for adaption of CAR-T cells, described as hypoxic, acidic, and nutrient-deprived [
21–
23]. Previous studies demonstrated that effector T cells rely on glucose metabolism [
24], and low levels of glucose impair T cell function through regulating mTOR pathway [
25]. Different from effector T cells, Treg cells can utilize substrates either derived from glycolysis or fatty acid oxidation for expansion [
26]. Accordingly, hypoxia and fatty acid metabolic pathways were highly expressed in effector T cells and Treg cells from the progressed patient, whereas glycolytic-associated pathways were upregulated in the CR patient, indicating that the hypoxic microenvironments may alter T cell metabolism at disease progression to favor the thriving of the immunosuppressive Treg cells. On the other hand, the T subtype compositions in CAR-T cell products might be different between PR/PD and CR patients. In a previous report, the relative proportions of central memory CD8
+ T cells and exhausted CD8
+ T cells in CAR-T cell products predicted the therapeutic outcome of CAR-T cells. Patients infused CAR-T cells with higher percentages of central memory CD8
+ T cells were more likely to benefit from CAR-T treatment [
27]. The prognostic value of T subtype compositions in CAR-T products will be given attention in our future project. We also found that tumor-infiltrating T cells were exhausted in the progressed patient by expression of
LAG3,
TIGIT, and
PDCD1 on effector T cells. It was shown that T cell inhibitory receptors (such as PD-1 and CTLA-4) can decrease glucose uptake and inhibit the glycolytic processes, while PD-1 blockade reverses glucose restriction in tumor-infiltrating-T cells, enhancing CD8
+ T cell glucose influx and glycolysis via mTOR signaling [
28]. Therefore, these findings may provide insight into immunotherapeutic strategies to improve the clinical efficacy of CAR-T cell therapy.
Currently, this is, to the best of our knowledge, the first single-cell study to analyze the dynamic change of the tumor microenvironments in situ during CAR-T cell therapy. A pre-existing immunosuppressive tumor microenvironment inducing CAR-T cell dysfunction was identified and could be a useful tool for predicting clinical outcomes. Furthermore, our data provided a clinical rationale for targeting the tumor microenvironments and metabolic reprogramming to prevent lymphoma relapse after CAR-T cell therapy.
PDF
(7289KB)
